# Docs - Guides: Products and how-tos - **Get started** - [GPU VM quickstart](/docs/quickstart): From an API key to your first running GPU VM, with the orl CLI or plain curl: browse the catalog, add an SSH key, launch, connect, and stop billing. - [Authentication](/docs/authentication): Authenticate every request with an OpenRelay API key (or_…) sent as a Bearer token. Create, rotate, and revoke keys from the dashboard or the API. - [Permissions](/docs/permissions): Roles, scoped API keys, service accounts, guardrails and Cedar policies decide what each member and key can do. - **Compute** - [VM billing and burn rate](/docs/vm-billing): How VM billing works: one bundled hourly rate per VM, locked when it starts, CPU and GPU sizing, and how long your balance lasts before auto-stop. - [SSH troubleshooting](/docs/ssh-troubleshooting): Fix Permission denied (publickey), "No running VM found" and other SSH errors on an OpenRelay GPU VM, verify its host key, and connect from Windows. - [Endpoint authentication](/docs/endpoints): A private VM or Pod exposes an HTTPS endpoint. Authenticate to it with your OpenRelay API key sent as a Bearer token. - [Pods and persistent storage](/docs/pods): Bring-your-own-container GPU pods, what survives a stop, restart, crash or node loss, and how the /workspace volume is billed and retained. - [Managed Kubernetes](/docs/kubernetes): Kubernetes clusters whose control plane we run, with whole GPU or CPU hosts as nodes. Create a cluster and a node pool, write a kubeconfig, and use kubectl. - **Inference** - [Overview](/docs/inference): The OpenRelay Inference API, with OpenAI-compatible chat completions and embeddings, an Anthropic-compatible Messages API, and typed decisions. - [Quickstart](/docs/inference/quickstart): From an API key to your first chat completion, unary and streaming, using nothing but curl against the OpenAI-compatible OpenRelay API. - [Authentication](/docs/inference/authentication): Authenticate every inference request with an OpenRelay API key (or_…), sent as a Bearer token or an x-api-key header. - [Models](/docs/inference/models): List available models with GET /v1/models and see the OpenRelay public model ids you can pass as the model field in an inference request. - [Using OpenAI SDKs](/docs/inference/openai-sdks): Use the official OpenAI Python and Node libraries with OpenRelay by setting base_url and api_key, including streaming and environment variables. - [Pricing & billing](/docs/inference/pricing): How inference is billed, a prepaid balance, per-token metering with distinct input, cached-input, and output rates, and the usage and cost on every response. - [Errors & rate limits](/docs/inference/errors): How the Inference API reports errors, the OpenAI-style error body, the status codes you'll see, and how to retry. - **Inference APIs** - [Chat Completions](/docs/inference/chat-completions): POST /v1/chat/completions, the OpenAI-compatible chat completions endpoint, its request parameters, and response shape. - [Messages (Anthropic)](/docs/inference/messages): POST /v1/messages, the Anthropic-compatible Messages endpoint, its request parameters, response shape, and cache-token billing. - [Responses (OpenAI)](/docs/inference/responses): POST /v1/responses, the OpenAI Responses endpoint in stateless mode, the fields it refuses, its typed streaming events, and how responses are billed. - [Codex CLI](/docs/inference/codex): Point the OpenAI Codex CLI at OpenRelay through the Responses endpoint, with a config.toml provider block and an OpenRelay API key. - [Embeddings](/docs/inference/embeddings): POST /v1/embeddings, the OpenAI-compatible embeddings endpoint, its request and response shapes, limits, and input-only billing. - [Decisions](/docs/inference/decisions): POST /v1/decisions answers typed questions (choice, score, yes/no) about one input, with probabilities per question. Shapes, limits, billing, errors. - [Streaming](/docs/inference/streaming): Stream chat completions as Server-Sent Events with stream:true, the SSE format, the [DONE] sentinel, and usage and cost in the final chunk. - Batch API - [Overview](/docs/inference/batch): Submit chat completions, text completions, or embeddings as a JSONL file, run them within 24 hours, and download the results at 50% off the Sync rate. - [Quickstart](/docs/inference/batch/quickstart): Run a batch end to end. Build a JSONL file, upload it, create a batch, poll to completion, and download the results, with curl, the orl CLI, or the OpenAI SDK. - [Files](/docs/inference/batch/files): Upload batch input files as JSONL (a direct multipart upload for smaller files, a presigned URL for large ones), then read file metadata and content. - [Batches](/docs/inference/batch/batches): Create, get, list, and cancel batches. The batch object, inline vs file input, the status lifecycle, and pagination. - [Results](/docs/inference/batch/results): The Batch API output and error file formats: one JSONL line per request, joined back to your inputs by custom_id, and how to reconcile usage. - [Workflows](/docs/inference/batch/workflows): Run dependent inference inside one batch record. Your callback returns the next requests and OpenRelay runs them, bounded by steps, requests, and tokens. - Async API - [Overview](/docs/inference/async): Submit inference requests as a job, delivered within 1 hour or 24 hours at a discount from Sync. Read results as they finish or get them by webhook. - [Quickstart](/docs/inference/async/quickstart): Create an async job with curl, the orl CLI or the OpenAI Python SDK, watch it finish, and read the results of the requests it ran. - [Jobs](/docs/inference/async/jobs): The async job object, its statuses and counts, how to create, list, get and cancel one, and how to retry a create safely. - [Results](/docs/inference/async/results): The shape of an async result, how to page through the results of a job while it runs and after it ends, and why there is no output file. - [Delivery webhooks](/docs/inference/async/webhooks): Get an async job's results pushed to your endpoint when it ends, with the three async.job events, the payload, signing and retries. - [Agents](/docs/inference/async/agents): What an async request can do for an agent today. Tool calls come back for your own loop; OpenRelay-run tool loops and workflows are not available in Async. - [Limits and errors](/docs/inference/async/limits-and-errors): The size limits of an async job, the errors a create can answer with, and the codes a failed request carries in its result. - [Pricing](/docs/inference/async/pricing): Within 1 hour is 20% off Sync, within 24 hours 50% off. Only completed requests are charged; one not started by the deadline expires unbilled. - [Dedicated endpoints](/docs/inference/dedicated): Serve a model on capacity reserved for your organization, call it with its private model id, and monitor its requests, errors, latency, and tokens. - **Providers** - [Become a provider](/docs/providers): Bring your GPU hardware online as OpenRelay capacity with a single command and start earning. Covers provisioning tokens, node security, and earnings. - [Provider nodes](/docs/provider-nodes): Hardware requirements for VM and pod nodes, GPU setup, health checks, and the full lifecycle of your provider hardware. - [Interruptible nodes](/docs/interruptible-nodes): Onboard a GPU container you rent from another market or cloud as Interruptible capacity with one command, and tell OpenRelay when the market takes it back. - API Reference: Every REST endpoint - **Using the API** - [Introduction](/docs): The OpenRelay REST API, deploy GPU VMs and pods, manage organizations and billing, and automate your infrastructure. - [Errors](/docs/errors): How the OpenRelay API reports errors: the JSON error body, status codes, when to retry, idempotent creates, and capacity and balance refusals. - [Pagination](/docs/pagination): How to page through OpenRelay list endpoints with cursors: the request parameters, the response shape, and a loop that iterates every page. - **Account and access** - Account: The signed-in user: profile, memberships, and first-time onboarding, plus the invites addressed to you and the identity and org each request acts in. - [Overview](/docs/account/overview): The signed-in user: profile, memberships, and first-time onboarding, plus the invites addressed to you and the identity and org each request acts in. - [Create your profile and first org (idempotent)](/docs/account/bootstrap): Creates the signed-in user's profile and a first organization they own (named orgName, or My Organization). A user already set up gets their default org back. Requires a signed-in dashboard session token; API keys cannot call this endpoint. - [Current user profile and org memberships](/docs/account/getMe): Returns the signed-in user's profile and organization memberships. Requires a dashboard session token; API keys cannot call this endpoint. - [List pending invites to the signed-in user](/docs/account/listMyInvites): Returns the unexpired pending invites whose invited email matches the session's email, with each organization's name, the role offered and the inviter's email. Requires a dashboard session token; API keys are organization keys and cannot hold or accept invites. - [Accept a pending org invite (session only)](/docs/account/acceptOrgInvite): Creates the membership for the invite addressed to the session's email in the given org, then consumes the invite. Returns the org id and the role granted. This is the only way an invite becomes a membership: the invitee must be authenticated with the invited email and act explicitly. - [Decline a pending org invite (session only)](/docs/account/declineOrgInvite): Deletes the pending invite to the given organization addressed to the session's email, without joining. Returns 204 whether or not an invite existed. - [The current principal and the org it acts in](/docs/account/whoami): Returns the calling principal (API key or user session) and the organization the request acts in, with the org's name and, for a key, its id and name. For an API key that org is fixed by the key; for a session it is the active org, and availableOrgs lists every org the user belongs to. Unlike /v1/me this works for API keys and does not onboard. - [Update the caller's profile](/docs/account/updateProfile): Sets the signed-in user's full name, avatar URL or default organization and returns the profile. The default must be an org the user belongs to (403 otherwise). Requires a signed-in dashboard session token; API keys cannot call this endpoint. - [Start a CLI device login (no auth)](/docs/account/startCliLogin): Starts the orl auth login device flow: returns a device code for the CLI to poll and a 6-digit user code for a signed-in person to confirm. No API key or session is required: this is how an unauthenticated CLI starts the flow that gets it one. The code expires in 10 minutes. deviceName is the requesting host's best-effort hostname, shown on the approval page so the signed-in user can recognize their own terminal before confirming the code. - [Poll a device login for its session](/docs/account/pollCliLoginToken): Polls a device login: pending until the user approves, then returns the CLI session credential exactly once; denied or expired codes say so. RFC 8628-style polling: while the code is pending, answers 400 authorization_pending; faster than the returned interval, 429 slow_down; denied, 403 access_denied; expired or unknown, 410 expired_token. On the first poll after approval, mints and returns the one-time CLI session credential; every poll after that (the code was already collected) also answers expired_token, so the token is delivered to exactly one caller. - [Look up a pending device login](/docs/account/getCliLoginDeviceDetails): Shows the approval page what a pending device login is: the requesting device's name, the IP that started the flow and when. User sessions only. Approving or denying a CLI login is something only a signed-in person does in the dashboard, so neither an API key nor a CLI session can call this. Rate limited per user and per IP, the brute-force mitigation for a guessed 6-digit code. - [Approve a device login](/docs/account/approveCliLogin): Approves a pending device login, giving the CLI a 30-day session as you in the chosen org, acting with the role you pick. Needs a signed-in user session. The session acts as you, so it can do what the dashboard can, limited to the role it was given: every request is decided with that role and with your role in the org, and allowed only when both allow it, so demoting you narrows the session at once. The roles it can be given are the ones GET /v1/me/cli-session-roles offers; without roleId it gets the default that operation names. - [Deny a pending device login (session only)](/docs/account/denyCliLogin): Marks the code denied; the polling CLI gets access_denied and stops. Needs a signed-in dashboard session: a CLI session cannot call this. Irreversible for that code (a fresh `orl auth login` starts a new one). - [List roles for a CLI session](/docs/account/listCliSessionRoles): The roles the approval page offers a CLI session in an org, and the one it starts with. Needs a signed-in user session. Each built-in role up to Admin, and each custom role, that you could give a member yourself (what its holder reaches is within what you can do), and always your own role unless it is Owner. defaultRoleId is Member when offered, otherwise your own role. - [List your CLI sessions, every org (session only)](/docs/account/listCliSessions): Lists every orl auth login session you hold across your orgs, with device name, role, creation, last use and expiry. Needs a signed-in user session. Sessions that expired or were revoked are listed too. An API key cannot call this: it cannot enumerate the user sessions that out-rank it. - [Revoke one of your CLI sessions (session only)](/docs/account/revokeCliSession): Revokes one of your CLI sessions; it stops working on the next request to the API or the inference gateway. orl auth logout calls this. Both the control plane and the inference gateway check revocation on every request, with no cache to wait out. Revoking an already revoked session answers 204 again; orl auth logout then deletes the credential locally. - Organizations: Organizations and their members: create, update or soft-delete an org, invite, remove or re-role members, or leave one. Most resources are scoped to an org. - [Overview](/docs/organizations/overview): Organizations and their members: create, update or soft-delete an org, invite, remove or re-role members, or leave one. Most resources are scoped to an org. - [List an org's members](/docs/organizations/listOrgMembers): Returns every member of the organization with their user id and role, plus the email, full name and avatar from their profile and their invite and join times. - [List an org's pending invites (owner/admin)](/docs/organizations/listOrgInvites): Returns the organization's invites not yet accepted, declined or revoked, each with the email, role, inviter's user id and expiry. Expired invites are still listed. - [Revoke a pending invite (owner/admin)](/docs/organizations/revokeOrgInvite): Deletes the pending invite for an email address so it can no longer be accepted. Returns 204, including when no invite exists; the email match ignores case. - [Record the caller's signed-in browser device](/docs/organizations/recordOrgDevice): Records the Fingerprint visitor id of the browser a member is signed in from against an existing organization, and checks the device like a sign-in. It covers organizations created automatically at first sign-in, whose creation carries no device. Any member of the org may call it. Under an enforcing device policy, a device that fails the check places the org on a verification hold. It always returns 204 once authorized, including when the body is empty: a browser that blocked the agent must not see an error. - [Create an organization (caller becomes owner)](/docs/organizations/createOrg): Creates an organization owned by the caller and returns its id and the owner role. A custom slug must be unique; one account owns at most five organizations. Requires a signed-in dashboard session token; API keys cannot call this endpoint. - [Get an organization](/docs/organizations/getOrg): Returns the organization's ids, name, slug, plan, billing email, tax id, invoice address and account status (active, pending, restricted, suspended or banned). - [Update an organization (owner/admin)](/docs/organizations/updateOrg): Changes the organization's name, billing email, tax id and country, and invoice address, and returns it. An empty string clears the tax and address fields. - [Soft-delete an organization (owner only)](/docs/organizations/deleteOrg): Owner-only soft-delete: members lose access and the org's remaining workloads are stopped. Support can restore the org, but not its Kubernetes clusters. Every cluster is deleted with the org, its workloads and data removed without a drain, and so is every provider node of the org that has stopped taking new work (delisted or under a leave notice) and hosts nothing. 409 while any VM other than a Kubernetes cluster's worker is active (HAS_ACTIVE_VMS), the balance is not zero (NONZERO_BALANCE), a provider node still takes new work or hosts workloads (PROVIDER_NODE_ACTIVE), or provider earnings are not yet paid out (UNPAID_PROVIDER_EARNINGS). - [Invite a member by email (owner/admin)](/docs/organizations/addOrgMember): Invites an email address to the organization with any role but owner, up to your own access (custom roles need permissions enabled). Expires in seven days. Records a pending invite and emails the address. The response status is always "invited"; the user becomes a member only after they accept (POST /v1/me/invites/{orgId}/accept). Re-inviting the same email refreshes the invite's role and expiry. 409 if the email already belongs to a member. - [Leave an organization](/docs/organizations/leaveOrgMember): Removes the signed-in user from the organization and revokes their playground keys. Organization keys are not affected. The organization must retain at least one owner. - [Update a member's role (owner/admin)](/docs/organizations/updateOrgMemberRole): Sets a member's role and returns the updated membership. Custom roles need permissions enabled. Setting owner adds an owner, and only an owner can grant it. - [Remove a member (owner/admin)](/docs/organizations/removeOrgMember): Removes a member from the organization so they lose access to it. With permissions enabled, their personal keys are revoked and the response lists them. Without permissions enabled the response is 204. - [Add a note to a live verification hold](/docs/organizations/createVerificationNote): Adds a statement (company, use case, link) to the org's live verification hold for the reviewer; it gives context and never lifts the hold itself. Answers 409 when the org has no active hold, or already holds the maximum number of notes. - API Keys: Programmatic access keys (or_…; legacy vl_… still valid) used to authenticate with this API. The plaintext is returned once, when the key is created. - [Overview](/docs/api-keys/overview): Programmatic access keys (or_…; legacy vl_… still valid) used to authenticate with this API. The plaintext is returned once, when the key is created. - [List an org's API keys (no secrets)](/docs/api-keys/listOrgApiKeys): Returns the organization's API keys with each key's id, name, prefix, kind, scopes, expiry and last use. The plaintext secret is never returned. With permissions enabled, a caller without apikey.read gets only their own personal keys. Dashboard playground keys are never listed. - [Create an API key (returns the plaintext once)](/docs/api-keys/createOrgApiKey): Returns the new key's plaintext secret, which is shown only in this response, with its id, prefix, organization id, and any permissions. With permissions enabled, only a signed-in session creates keys, and a new key belongs to the organization (kind org), so it keeps working if its creator leaves; personal and service kinds are refused, except dashboard playground keys, which are always personal. Its permissions come from template or permissions and can never exceed the creator's own access (403 ESCALATION_REFUSED). scopes still works and maps to the key's ceiling. - [Revoke an API key](/docs/api-keys/revokeOrgApiKey): Revokes an API key in the organization, after which requests made with it no longer authenticate. With permissions enabled an unknown key id returns 404. - Roles: Built-in and custom roles. A custom role is a set of permissions plus optional model, endpoint and network conditions. Available to organizations with permissions enabled; otherwise these endpoints return 404. - [Overview](/docs/roles/overview): Built-in and custom roles. A custom role is a set of permissions plus optional model, endpoint and network conditions. Available to organizations with permissions enabled; otherwise these endpoints return 404. - [List built-in and custom roles](/docs/roles/listRoles): Returns the built-in roles and the organization's custom roles, each with its permission ids and how many members hold it. The retired inference role is listed only while someone still holds it. - [Create a custom role](/docs/roles/createRole): Creates a custom role from a set of permissions plus optional model, endpoint and network conditions, and returns it. A role cannot grant more than you hold. Requires a signed-in session. A role that grants more than its creator holds is refused with 403 ESCALATION_REFUSED, which names the first missing permission. - [Get a role, with its generated Cedar](/docs/roles/getRole): Returns one built-in or custom role with its permission ids and holder counts. A custom role also carries its conditions and the Cedar generated from it. - [Update a custom role](/docs/roles/updateRole): Changes a custom role's name, description, permissions, or its model, endpoint and network conditions, and returns it. Built-in roles cannot be changed. Requires a signed-in session. The same no-escalation rule as creation applies. - [Delete a custom role](/docs/roles/deleteRole): Deletes one of the organization's custom roles and returns 204. A role still held by a member is refused with 409 ROLE_IN_USE. Requires a signed-in session. - Access: The permission catalog, effective access, the access checker, guardrails, Cedar policies and key settings such as the maximum key lifetime. Available to organizations with permissions enabled; otherwise these endpoints return 404. - [Overview](/docs/access/overview): The permission catalog, effective access, the access checker, guardrails, Cedar policies and key settings such as the maximum key lifetime. Available to organizations with permissions enabled; otherwise these endpoints return 404. - [Get the permission catalog](/docs/access/getPermissionCatalog): Returns the permission areas and levels with the actions each allows, the built-in roles, the key templates, and the actions API key ceilings cannot grant. - [Show what a principal can do right now](/docs/access/getEffectiveAccess): Returns the permission ids and actions a member or API key holds right now, and the models it may call, after its restrictions apply. Without parameters, answers for the caller. Naming a member or key of the organization needs policy.read. The answer is the principal's role intersected with its key restrictions, role conditions and the organization's guardrails. Network rules are checked against the address of the request you send, not one the named principal would call from. - [Check whether a principal may perform an action](/docs/access/checkAccess): Evaluates one action for a member or key of the organization and returns the decision, the layer that decided it and the deciding policies. - [List the org's guardrails and custom policies](/docs/access/listPolicies): Returns every guardrail and custom Cedar policy in the organization, enabled or disabled, with its networks, models, endpoints and Cedar text. - [Create a guardrail or a custom Cedar policy](/docs/access/createPolicy): Saves a guardrail (key networks, everyone networks or a model allowlist) or a custom Cedar policy and returns it, enabled unless you send enabled: false. Requires a signed-in session. Guardrails that apply to API keys only need guardrail.manage; the everyone_networks guardrail and every custom policy can block sessions and need policy.manage (owners). A policy that would take policy.manage or member.manage from the saving owner is refused with 409 WOULD_LOCK_OUT. - [Validate a policy without saving it](/docs/access/validatePolicy): Returns line and column diagnostics and the lockout check for a candidate policy. Requires a signed-in session and the permission saving it would need. - [Get a guardrail or custom policy](/docs/access/getPolicy): Returns one guardrail or custom policy by id: its enabled state, its networks, models or endpoints, and its Cedar (the source, or what a guardrail generates). - [Update, enable or disable a policy](/docs/access/updatePolicy): Changes a policy's name, description, networks, models, endpoints, Cedar text or enabled state, and returns it. Fields left out keep their values. Requires a signed-in session and the permission creating the policy needs. - [Delete a policy](/docs/access/deletePolicy): Deletes a guardrail or custom policy and returns 204. A delete is refused with 403 when removing the policy would let the organization do what you cannot. Requires a signed-in session and the permission creating the policy needs. - [Get the organization's key settings](/docs/access/getKeySettings): Returns the organization's maximum API key lifetime in days (0 means keys may never expire), with who last changed it and when. - [Set the maximum key lifetime](/docs/access/updateKeySettings): Sets maxKeyLifetimeDays, from 0 to 3650, and returns the saved settings. New API keys must expire within that many days; 0 lets keys never expire. Requires a signed-in session. - [Request access to a product](/docs/access/createAccessRequest): Submits a request for access to a product, from the dashboard's Request access flow. An operator reviews it in the ops queue. - Audit Events: The organization's access audit log. Available to organizations with permissions enabled; otherwise this endpoint returns 404. - [Overview](/docs/audit-events/overview): The organization's access audit log. Available to organizations with permissions enabled; otherwise this endpoint returns 404. - [List the org's access audit log, newest first](/docs/audit-events/listAuditEvents): Every change to keys, members, roles and policies, kept for 365 days. Filter by actor, event, target or time; 50 per page by default. - Transfers: Move resources between organizations you belong to. A transfer offers one VM; the receiving org can accept it within 7 days or reject it. The sender can cancel it while it is pending. - [Overview](/docs/transfers/overview): Move resources between organizations you belong to. A transfer offers one VM; the receiving org can accept it within 7 days or reject it. The sender can cancel it while it is pending. - [List orgs you can transfer resources to](/docs/transfers/listTransferTargetOrgs): The organizations the caller belongs to, other than the org the request acts in, sorted by name with id, name and slug, for choosing a transfer target. Requires a signed-in user session; a request without one returns 401. - [Start a resource transfer (owner/admin)](/docs/transfers/initiateTransfer): Offers one of the org's VMs to another organization, named by targetOrgSlug, and returns the pending transfer, which can be accepted for 7 days. A VM can have one unexpired pending transfer at a time; a second returns 409, as does an inference serving replica. Naming the caller's own org returns 400. An organization under review or being verified cannot offer a VM: the offer answers 403 ACCOUNT_RESTRICTED or ACCOUNT_PENDING_VERIFICATION. - [Pending incoming transfers (enriched)](/docs/transfers/listIncomingTransfers): The pending transfers addressed to the org, newest first, each with the VM's name (Deleted VM once it is gone) and the sending org's name and slug. A transfer past its 7-day expiry stays listed as pending until it is rejected or cancelled, or an accept attempt marks it expired. - [Outgoing transfers, all statuses (enriched)](/docs/transfers/listOutgoingTransfers): The transfers the org has sent, in every status, newest first and at most 50, each with the VM's name and the receiving org's name and slug. - [Count of pending incoming transfers](/docs/transfers/getTransferPendingCount): The number of transfers addressed to the org that are still pending, returned as count: the same transfers the incoming list returns. It includes transfers past their 7-day expiry that have not yet been accepted, rejected or cancelled. - [Cancel a pending outgoing transfer (source org)](/docs/transfers/cancelTransfer): Withdraws a pending transfer the caller's org sent and marks it cancelled. A transfer that is not pending, or that this org did not send, returns 404. - [Accept an incoming transfer (owner/admin)](/docs/transfers/acceptTransfer): Moves the offered VM to the caller's org and marks the transfer completed. At least one sshKeyIds entry is required; those keys replace the source org's. The source org's GitHub runner config for the VM is deleted, the LLM keys it holds for the VM are revoked, and any other pending transfer of the VM is cancelled. An expired transfer returns 400. If the VM has since moved to a third org, the transfer is cancelled and the call returns 409. The accepting organization takes the VM on as a new workload: one under review, being verified, or whose available balance is under the minimum answers 403 ACCOUNT_RESTRICTED, ACCOUNT_PENDING_VERIFICATION or INSUFFICIENT_BALANCE. When the sending organization can no longer hand the VM over, the call answers 409 TRANSFER_UNAVAILABLE. When either organization's account changes while the accept runs, it answers 409 TRANSFER_ACCOUNT_CHANGED; try again. Each of these leaves the transfer pending, so it can still be rejected, or accepted later. - [Reject an incoming transfer (owner/admin)](/docs/transfers/rejectTransfer): Declines a pending transfer addressed to the caller's org and marks it rejected; the VM stays with the org that offered it. Returns 204 with no body. - **Compute** - VMs: GPU virtual machines: create, stop, restart or terminate a VM, manage its disks and SSH keys, and read its live burn rate and runway. - [Overview](/docs/vms/overview): GPU virtual machines: create, stop, restart or terminate a VM, manage its disks and SSH keys, and read its live burn rate and runway. - [List an org's VMs (cursor-paginated)](/docs/vms/listOrgVms): Returns the org's VMs as a cursor-paginated page of VmSummary items, without pods (listPods returns those); set activeOnly to leave out terminated VMs. The nodes of the org's managed Kubernetes clusters are not listed either; listKubernetesNodes returns those. - [Get a VM by id](/docs/vms/getVm): Returns one VM or pod of the caller's organization as a VmSummary: its status and reason, GPU model and count, endpoint URL, visibility and hourly price. - [Get a VM with GPUs, location, template and price](/docs/vms/getVmDetail): Returns a VM's full record: its shape and disk, SSH connection details and readiness, GPU model and VRAM, location, OS template, and its hourly rate. - [Get a VM's burn rate, session cost and runway](/docs/vms/getVmBurn): Returns a VM's bundled hourly rate, the cost its open session has accrued, and the org's available balance, total burn rate and projected runway. Every VM bills at one bundled hourly rate (per-GPU catalog price times GPU count for GPU VMs, a flat size rate for CPU VMs); disk is included, never billed separately. The rate is locked when the usage session opens, so this endpoint reads it from the open usage record when one exists (rateSource=locked) and only falls back to the current catalog when nothing is accruing (rateSource=catalog). - [Pod GPU metrics](/docs/vms/getVmMetrics): Returns utilization, memory, power and temperature time series for each of a pod's GPUs over the selected range, sampled on the host that runs the pod. The telemetry is host-side (nvidia-smi on NVIDIA hosts, amd-smi on AMD), sampled on the node that runs the pod and returned as one continuous time grid. Every series and every GPU in a response shares that grid: the value at index i of any values array was measured at start plus i times stepSeconds, and a null means the node reported no sample at that point. The window ends now and covers only the periods this pod actually held the GPUs it is reported against, so a card that ran another customer's pod earlier never contributes a sample here. A GPU attached to a VM is passed through to the guest, so the host cannot read it. That case answers 200 with an accelerator status of unavailable and an empty gpus list rather than an error, and so do a pod that has not started yet, a pod that was not running in the selected range, and a metrics store that cannot be reached: the accelerator object is the answer, and the client renders its detail sentence. A host that stopped reporting while the pod is still running answers degraded with the history up to the last sample still attached. This operation is exposed as an agent tool by default, in the same class as a pod's burn rate: "how busy were my GPUs" is a read-only question about a resource the caller already owns. - [Create a VM](/docs/vms/createVm): Creates a GPU or CPU VM from the requested shape, image, disk and SSH keys and returns it as a VmSummary; a valid request with no free capacity answers 503. - [Stop a VM](/docs/vms/stopVm): Stops the workload and releases its GPU. A VM keeps its disk; a pod keeps its /workspace volume and its node, and can be started again with restart. While stopping VMs is switched off for the organization, a VM stop answers 409 STOP_DISABLED: terminate the VM, or leave it running. Pods are not affected. - [Restart a VM](/docs/vms/restartVm): Reboots a running VM in place, or starts a stopped or failed VM or pod again on the node that holds its disk or volume; a running pod's container is recreated. A stopped or failed VM answers 503 VM_NODE_BUSY when its node has too few free GPUs of the VM's model, a GPU it needs has failed, or too little disk, RAM or CPU, to take it back, and the message names which (a failed GPU by its address); try later, or create a new VM to place elsewhere. A running VM or pod is never rebooted on a GPU that has failed: it answers 503 VM_NODE_BUSY or POD_NODE_BUSY naming the failed GPU, and the message says when stopping it and starting it again would place it on healthy GPUs of its node. A stopped or failed pod re-claims its GPUs on that node (503 POD_NODE_BUSY when that node's GPUs are taken, 503 INSUFFICIENT_IP_CAPACITY when the pod's InfiniBand fabric has no free address; 409 POD_NODE_OFFLINE when the node is offline, POD_NODE_NO_INFINIBAND when the org is enabled for InfiniBand but the node's agent predates it, POD_NODE_NO_AMD_GRANT when the pod's GPU model is AMD but the node's agent is below the level that grants AMD GPUs, POD_STOPPING while the previous run is still tearing down, with Retry-After and a message naming the wait until it is expected to finish). Recreating a running pod's container resets its filesystem and keeps /workspace. - [Terminate a VM](/docs/vms/terminateVm): Terminates a VM or pod for good: its workload stops, its GPUs are released, its usage session closes and its disk is deleted. A repeat terminate answers 200. - [Reboot a VM (beta, not yet executed)](/docs/vms/rebootVm): Beta: the reboot request is validated and accepted, but the in-guest reboot is not yet executed by the data plane. - [Set a VM's endpoint visibility](/docs/vms/setVmVisibility): Makes a VM's HTTP endpoint public (no data-plane auth) or private (the org's API key required) and returns the VM; a running VM's route changes at once. - [Resize a stopped VM's disk](/docs/vms/resizeVmDisk): Beta: a stopped VM's new disk size is validated and recorded, but the disk is not yet resized on the node. The disk can grow, never shrink. A pod's volume cannot be resized. - [SSH keys attached to a VM + the org's keys](/docs/vms/getVmSshKeys): Lists every SSH key of the VM's organization with its name and fingerprint, each flagged attached when it is in this VM's allowed set of SSH keys. - [Attach an org SSH key to a VM](/docs/vms/attachVmSshKey): Adds one of the org's SSH keys to the VM's allowed set, so the gateway accepts that key for SSH into the guest. Attaching a key already attached answers 204. The key must be a live key of the VM's organization, and a terminated VM answers 400. - [Detach an org SSH key from a VM](/docs/vms/detachVmSshKey): Removes the key from the VM's allowed set, so the gateway stops accepting it for SSH into the guest, and answers 204, also when the key was not attached. The last attached key cannot be detached: the guest would have no way in over SSH, so that returns 400. - [Get a VM's console log tail](/docs/vms/getVmLogs): Returns the tail of a VM's serial console, capped to a bounded size; a VM with no logs yet answers 200 with found false and a message, not an error. A transient state (no allocation yet, logs temporarily unavailable) answers 200 with found false, so a polling UI renders a state instead of an error. The cap is the one pod logs use, so a long-running console never exceeds the response size orl reads. Also works for pods: `orl vms logs` and `orl pods logs` are the same operation pair with different id params, and `orl vms ` reaches every other pod lifecycle command. - Pods: GPU pods: bring-your-own-image containers on whole GPUs, with SSH access and an HTTP endpoint. Create and list pods, and check live capacity and rates. - [Overview](/docs/pods/overview): GPU pods: bring-your-own-image containers on whole GPUs, with SSH access and an HTTP endpoint. Create and list pods, and check live capacity and rates. - [List an org's pods (cursor-paginated)](/docs/pods/listPods): Returns the live compute-fabric pods of the org the API key resolves to as a cursor-paginated page of VmSummary items; terminated pods are left out. A pod is a managed_by=fabric Vm row (capabilityTier=pod), so each item is a VmSummary and paging mirrors listOrgVms. - [Create a pod](/docs/pods/createPod): Runs your container image as a pod on gpuCount whole GPUs of one node, with SSH access, at the on-demand rate fixed at creation, and returns it as a VmSummary. Compute-fabric pod (BYO container image + SSH keys): a hardened GPU container on gpuCount whole native-driver GPUs of one node. Admission claims free device-request GPU units (see listPodCapacity for what is claimable), snapshots the on-demand rate immutably onto the pod, and drives the guest workflow. A valid request the fleet has no room for right now answers 503 with a capacity code; a GPU count no node could ever hold for the org answers 400 GPU_COUNT_NOT_ALLOCATABLE; and an org without the pod tier answers 409 FEATURE_DISABLED. - [Live pod GPU capacity and on-demand rates](/docs/pods/listPodCapacity): Returns, per GPU model, the GPU units a pod create can claim right now, the largest single pod that fits, and the server-computed on-demand rate per GPU-hour. What a pod create can actually claim right now, per GPU model: free native (device-request) GPU units on online, non-delisted nodes a pod can be sized on (the exact filter admission uses), plus the SERVER-computed on-demand hourly rate. Only models with pod units and a price are returned; a create flow should offer exactly these and render these rates verbatim rather than re-deriving prices from the catalog. - [Get a pod's container log tail](/docs/pods/getPodLogs): Returns the tail of a pod's container output (startup, CUDA errors, a crash's last lines), capped to a bounded size, or found false while it starts. - Kubernetes: Managed Kubernetes: clusters whose control plane we run, with whole GPU or CPU hosts as nodes. Create clusters, scale node pools, and get kubeconfigs. Managed Kubernetes is in limited availability and is enabled per organization; without it every operation answers 409 FEATURE_DISABLED. Every operation acts in the caller's own organization: the API key's, or the session's X-Org-Id. The server checks every request against the parameters and request bodies documented here and answers 400 BAD_REQUEST naming the violation, so each constraint is enforced as written. A 409 INVALID_CLUSTER_STATE names the cluster status that blocks the operation. - [Overview](/docs/kubernetes/overview): Managed Kubernetes: clusters whose control plane we run, with whole GPU or CPU hosts as nodes. Create clusters, scale node pools, and get kubeconfigs. Managed Kubernetes is in limited availability and is enabled per organization; without it every operation answers 409 FEATURE_DISABLED. Every operation acts in the caller's own organization: the API key's, or the session's X-Org-Id. The server checks every request against the parameters and request bodies documented here and answers 400 BAD_REQUEST naming the violation, so each constraint is enforced as written. A 409 INVALID_CLUSTER_STATE names the cluster status that blocks the operation. - [List Kubernetes clusters](/docs/kubernetes/listKubernetesClusters): The organization's managed Kubernetes clusters, newest first, in cursor pages of summaries with each cluster's status, health, and node and Ready counts. A legacy API key needs clusters:read. - [Create a Kubernetes cluster](/docs/kubernetes/createKubernetesCluster): Starts creating a cluster with a managed control plane in one data center and answers 202 with its id; poll getKubernetesCluster for its status. Node pools in the request are added once the control plane is ready; add more later with createKubernetesNodePool. A cluster's name is unique within the organization: a create with the name of one of its clusters answers 409 NAME_TAKEN until that cluster's deletion completes. An organization has a configured limit on clusters creating at once; a create at that limit answers 409 TOO_MANY_CLUSTERS_CREATING until one of them leaves creating. A create that meets a concurrent change to the organization answers 409 CONCURRENT_UPDATE; retry it. A create with a GPU node pool answers 503 VM_IMAGE_CATALOG_UNAVAILABLE while the control plane has not read the image catalog yet; retry it. A legacy API key needs clusters:write. - [Get a Kubernetes cluster](/docs/kubernetes/getKubernetesCluster): Returns one cluster: its status, Kubernetes version, data center, control-plane replicas, API endpoint, pod and service networks, node counts and conditions. A legacy API key needs clusters:read. - [Delete a Kubernetes cluster](/docs/kubernetes/deleteKubernetesCluster): Starts deleting the cluster, its node pools and every node, and answers 202; poll getKubernetesCluster, which answers 404 once deletion completes. Deletion is final: the cluster's state is not kept. Repeating the call finishes a deletion that could not remove everything. A legacy API key needs clusters:write. - [List a cluster's node pools](/docs/kubernetes/listKubernetesNodePools): Returns a cluster's node pools as a cursor-paginated page, each with its SKU, the node count it converges to, how many nodes are Ready, and its status. A legacy API key needs clusters:read. - [Add a node pool to a cluster](/docs/kubernetes/createKubernetesNodePool): Adds count whole-host nodes of one SKU, a GPU model or a whole-host CPU SKU, in the cluster's data center, and answers 202 with the new pool as creating. A cluster that is creating, deleting or failed answers 409 INVALID_CLUSTER_STATE; a failed cluster has no control plane to join. A GPU pool answers 503 VM_IMAGE_CATALOG_UNAVAILABLE while the control plane has not read the image catalog yet; retry it. A legacy API key needs clusters:write. - [Get a node pool](/docs/kubernetes/getKubernetesNodePool): Returns one node pool of the cluster: its name, its SKU (a GPU model or a whole-host CPU SKU), the node count it converges to, its Ready count and status. A legacy API key needs clusters:read. - [Delete a node pool](/docs/kubernetes/deleteKubernetesNodePool): Starts draining and removing every node in the pool, then the pool, and answers 202 with the pool's id and status deleting. A pool with no nodes is deleted at once. Poll getKubernetesNodePool, which answers 404 once the pool is removed. A cluster that is deleting answers 409 INVALID_CLUSTER_STATE. A legacy API key needs clusters:write. - [Scale a node pool](/docs/kubernetes/scaleKubernetesNodePool): Sets the pool's node count and answers 202 with the pool. Scaling down removes nodes and whatever runs on them, draining each one first. A cluster that is creating, deleting or failed answers 409 INVALID_CLUSTER_STATE; a failed cluster has no control plane to join. A legacy API key needs clusters:write. - [List a cluster's nodes](/docs/kubernetes/listKubernetesNodes): Returns a cluster's nodes as a cursor-paginated page, each with its pool, registration and Ready state, name once joined, public IP and timestamps. Filter to one pool with nodePoolId. A node's GPUs are on getKubernetesNode. A legacy API key needs clusters:read. - [Get a node](/docs/kubernetes/getKubernetesNode): Returns one node of the cluster: its registration and Ready state, its name in the cluster once it joins, the host's public IP and, on a GPU host, its GPUs. A legacy API key needs clusters:read. - [Delete a node](/docs/kubernetes/deleteKubernetesNode): Removes the node from the cluster and releases its host, draining it first unless drain is false, and drops its pool's count by one so it is not replaced. Answers 202 with the node's id and status removing. Poll getKubernetesNode, which answers 404 once removal completes. The pool's last node is not deleted: that answers 409 LAST_NODE_IN_POOL, and deleting the pool removes it. A node of a pool being deleted is removed with its pool: its delete is accepted and changes nothing. A cluster that is deleting answers 409 INVALID_CLUSTER_STATE. A legacy API key needs clusters:write. - [Mint a cluster token](/docs/kubernetes/createKubernetesClusterToken): Exchanges the caller's API key or session for a bearer token that only this cluster accepts, returned as an ExecCredential. The token expires after one hour. The token carries the requested role; mint another rather than storing it. The response is an ExecCredential so a kubectl exec credential plugin can print it as is. `orl kube token` is such a plugin, and `orl kube kubeconfig` writes a kubeconfig that runs it. A cluster that is deleting or failed answers 409 INVALID_CLUSTER_STATE. A restricted or pending organization can still mint. A legacy API key needs clusters:write. - [Get a kubeconfig](/docs/kubernetes/getKubernetesKubeconfig): Returns a kubeconfig for this cluster with a freshly minted one-hour token in the requested role embedded, the kind createKubernetesClusterToken returns. It holds no client certificate, and it stops working when the token expires after one hour. `orl kube kubeconfig` writes one that keeps working: it holds no token and runs `orl kube token` for each kubectl request. A cluster that is deleting or failed answers 409 INVALID_CLUSTER_STATE, and so does a cluster that is creating until its control plane's certificate authority exists. A restricted or pending organization can still get one. A legacy API key needs clusters:write. - Snapshots: Point-in-time VM snapshots and forking new VMs from them: snapshot a VM, list snapshots (optionally by source VM), soft-delete one, or fork a new VM from it. - [Overview](/docs/snapshots/overview): Point-in-time VM snapshots and forking new VMs from them: snapshot a VM, list snapshots (optionally by source VM), soft-delete one, or fork a new VM from it. - [Create a snapshot of a VM](/docs/snapshots/createVmSnapshot): Beta: the snapshot record is created and returned with status creating, but snapshot capture is not yet enabled on the data plane. A terminated VM answers 400, and a VM with a snapshot already being created answers 409. - [List VM snapshots, optionally by source VM](/docs/snapshots/listOrgSnapshots): Returns the org's VM snapshots as one array, or only one source VM's with vmId. Soft-deleted snapshots are included and carry deletedAt. - [Soft-delete a snapshot](/docs/snapshots/deleteSnapshot): Soft-deletes a snapshot: its status becomes deleting, deletedAt is set and the call answers 204. A snapshot that is already being deleted answers 409. - [Fork a new VM from a snapshot](/docs/snapshots/forkVm): Creates a VM from a snapshot, taking its GPU model and count, image, env vars, port and the source VM's SSH keys unless the request overrides them. The disk is never smaller than the snapshot's, and an unnamed fork is called fork-. A valid request with no free capacity answers 503. The response carries the new VM's id, name and status. A fork boots its snapshot's base image whatever shape it asks for. A fork of a GPU snapshot with gpuCount 0, or of a CPU snapshot with a gpuModelId and a gpuCount, is rejected with TEMPLATE_KIND_MISMATCH. A gpuModelId other than the snapshot's is admitted only when the template the snapshot's image was built from boots that model: otherwise it is rejected with TEMPLATE_NOT_COMPATIBLE, or with INVALID_TEMPLATE when the image catalog names no template for the image. While the control plane has not read the image catalog yet, such a fork answers 503 VM_IMAGE_CATALOG_UNAVAILABLE; retry it. - SSH Keys: Org-level SSH public keys that can be attached to VMs. Add a key and the server computes its fingerprint; list the org's keys or delete one. - [Overview](/docs/ssh-keys/overview): Org-level SSH public keys that can be attached to VMs. Add a key and the server computes its fingerprint; list the org's keys or delete one. - [List an org's SSH public keys](/docs/ssh-keys/listOrgSshKeys): Returns the org's SSH public keys with their fingerprints, how many live VMs each is attached to, and the VMs for which each is the only remaining key. - [Add an SSH public key (computes fingerprint)](/docs/ssh-keys/createOrgSshKey): Registers an OpenSSH public key for the org, stores it in canonical form with its computed fingerprint, and returns it with its new id and creation time. An invalid key answers 400 INVALID_KEY, and a key that is already registered answers 409 KEY_EXISTS. - [Delete an SSH public key](/docs/ssh-keys/deleteOrgSshKey): Deletes an SSH public key from the org and scrubs it from every VM it was attached to, unless it is the only key that still grants access to one of them. That case is refused with 409 (code KEY_IN_USE), since deleting it would strand that VM; attach another key to those VMs first. listOrgSshKeys reports this as blockingVms. - Registry Credentials: Private container-registry credentials for pulling images: basic auth, AWS ECR or GCP GCR. Secrets are stored encrypted and never returned. - [Overview](/docs/registry-credentials/overview): Private container-registry credentials for pulling images: basic auth, AWS ECR or GCP GCR. Secrets are stored encrypted and never returned. - [List an org's registry credentials (no secrets)](/docs/registry-credentials/listOrgRegistryCredentials): Returns the org's container-registry credentials with id, name, server, auth type and creation time. Stored passwords and keys are never returned. - [Create a registry credential](/docs/registry-credentials/createOrgRegistryCredential): Saves a registry login for the org: basic takes username and password; aws_ecr takes accessKeyId, secretAccessKey and region; gcp_gcr takes serviceAccountKey. A missing credential field, or an unknown authType, returns 400. The credentials are stored encrypted, and the response is the credential's summary without them. - [Delete a registry credential](/docs/registry-credentials/deleteOrgRegistryCredential): Removes a registry credential from the org by id and returns 204 with no body. Deleting an id that does not exist also returns 204. - Webhooks: Subscribe to platform events with signed HTTP callbacks for VM, batch, async job and balance events, and inspect each webhook's recent delivery attempts. - [Overview](/docs/webhooks/overview): Receive signed, retried HTTP callbacks for VM, batch, async job, balance and endpoint events, and verify each delivery's HMAC-SHA256 signature. - [List an org's webhooks (no secrets)](/docs/webhooks/listOrgWebhooks): Returns every webhook in the org with its URL, subscribed events, enabled flag and secret prefix. The full signing secret is shown only on create or regenerate. - [Create a webhook (secret returned once)](/docs/webhooks/createOrgWebhook): Registers an HTTPS URL for the events you list and returns the webhook with its whsec_ signing secret, the only time that secret is returned. Each delivery is a signed POST of {id, type, created_at, data}. Listing a batch event needs batch read access and a balance event billing read access, beside webhook write access (403 otherwise). A name the platform sends no event for is stored and never delivered. - [Update a webhook (name/url/events)](/docs/webhooks/updateOrgWebhook): Changes a webhook's name, URL or subscribed events and returns it. A field left out keeps its value; the signing secret and enabled state are not changed. Whatever it changes, an update needs the access the webhook's events need on create, since a new URL would redirect them. - [Delete a webhook](/docs/webhooks/deleteOrgWebhook): Deletes the webhook and returns 204. It gets no further deliveries, and any retries still pending for it are dropped instead of sent. - [Enable/disable a webhook](/docs/webhooks/toggleOrgWebhook): Sets enabled and returns the new state. A disabled webhook keeps its URL, events and secret but gets no deliveries. A retry that comes due while it is disabled is dropped. - [Regenerate a webhook secret (returned once)](/docs/webhooks/regenerateOrgWebhookSecret): Replaces the webhook's signing secret and returns the new whsec_ secret once. Later deliveries, retries included, are signed with the new secret. - [Recent delivery attempts for a webhook](/docs/webhooks/listOrgWebhookDeliveries): Lists a webhook's recent delivery attempts: event type, the id of the resource the event is about, attempt number, status, response code, error and time. The newest deliveries first, at most limit (default 50). A delivery of a batch event is left out for a caller without batch read access, and one of a balance event for a caller without billing read access, so a page can hold fewer than limit. - [VM and pod events](/docs/webhooks/vmEvent): Sent when a VM starts, or a VM or pod stops, is terminated, is interrupted or recovers, to each enabled webhook that lists the event or lists none. vm.running: a VM reached running from deploying, on its first start, on every later start of a stopped or failed VM and on every restart, a reboot of a running VM included; or a VM or pod recovered from an interruption. A pod's start, first or later, sends none. vm.stopped: a VM or pod was stopped. vm.terminated: a VM or pod was terminated. vm.interrupted: a VM or pod stopped serving through no action of yours, for example because its node went offline. It usually recovers on its own and sends vm.running again; one still interrupted after 7 days, or whose node is removed, is ended instead: a pod as terminated, which sends vm.terminated, and a VM as stopped, which sends no event. The events cover the VMs and pods your organization lists; the nodes of a managed Kubernetes cluster send none. Delivery is at least once: the same change can occasionally be recorded as a second event, with a new event id and a new delivery id. - [Batch events](/docs/webhooks/batchEvent): Sent once per batch when it reaches its final status, usually within a minute, to each enabled webhook that lists the event. The events are batch.completed, batch.failed, batch.expired (the 24 hour completion window ended first) and batch.cancelled. A batch that reached its final status more than 24 hours before it is picked up sends none. Listing a batch event needs batch read access. - [Async job events](/docs/webhooks/asyncEvent): Sent once per async job when it ends, usually within a minute, to each enabled webhook that lists the event, with the job and its results. The events are async.job.completed, async.job.cancelled and async.job.failed. A job whose deadline passed ends completed, with its unstarted requests counted expired, so there is no expired event. The body carries the job as GET /v1/async/jobs/{id} answers for it. When the job's results total 256 KiB or less and the create rejected no line, the body carries them in data.results and data.results_url is null; otherwise data.results is null and data.results_url is the job's results pages. A failed delivery is retried 7 times, after 10 seconds, 1 minute, 5 minutes, 30 minutes, 2 hours, 6 hours and 15 hours, the last about 23.6 hours after the first. Listing an async job event needs batch read access. - [Balance events](/docs/webhooks/balanceEvent): Sent when your available balance is $1.00 or less, or below $0.50, to each enabled webhook that lists the event, from a check about once a minute. balance.low: the available balance is between $0.50 and $1.00; it is sent again only after the balance has gone back above $2.00. balance.depleted: the available balance is below $0.50 and running workloads are being stopped. A balance that falls from above $1.00 to below $0.50 between two checks sends balance.depleted alone. The check covers an organization with a VM or pod running, with auto-recharge on, or with compute usage not yet billed. A balance-exempt organization gets neither event, and a check that starts an auto-recharge sends neither. Listing a balance event needs billing read access. - [Dedicated endpoint events](/docs/webhooks/endpointEvent): Sent when one of your dedicated inference endpoints changes state, to each enabled webhook that lists the event, usually within seconds of the change. endpoint.ready: the endpoint serves requests, after it was created or enabled. endpoint.disabled: the endpoint was disabled and its requests are refused; an endpoint deleted before this was sent sends it first. endpoint.deleted: the endpoint was deleted, its last event. endpoint.failed: the endpoint's model did not start, its replicas failing before they were ever ready on every hardware shape it runs on; its replicas are stopped, and the endpoint's stateReason says why. Listing an endpoint event needs endpoint read access. Delivery is at least once: the same change can occasionally arrive twice, with the same event id. Each delivery is attempted on its own, so events can arrive out of order: order an endpoint's events by data.at. - Runners: Managed GitHub Actions runner pools: connect a GitHub installation, create, resize or delete pools, and list each pool's jobs, metrics and cost comparison. - [Overview](/docs/runners/overview): Managed GitHub Actions runner pools: connect a GitHub installation, create, resize or delete pools, and list each pool's jobs, metrics and cost comparison. - [List runner pools (with active/queued usage)](/docs/runners/listRunnerPools): Returns the organization's GitHub runner pools, newest first, with active and queued runner counts and the RAM they hold. Deleted pools are left out. - [Create a runner pool](/docs/runners/createRunnerPool): Connects a GitHub organization's App installation as a runner pool and starts its subscription. Repeating it returns the pool while it is active or suspended. name defaults to the GitHub owner, poolRamGb to 64 when omitted or 0, and diskSizeGb to 100 when omitted. A payment method must be on file (CARD_REQUIRED), and a personal GitHub account is refused (USER_ACCOUNT_UNSUPPORTED). When the OAuth flow is configured, pass the installation's grant from listRunnerInstallations (GRANT_REQUIRED). - [Get the GitHub connect URL](/docs/runners/getRunnerConnectUrl): Returns the URL that starts a GitHub connection: the OAuth authorize URL when OAuth is configured, otherwise the GitHub App install URL. installUrl carries the App install URL (empty when no App is configured), for installing the App on another GitHub organization. - [List GitHub installations you can connect](/docs/runners/listRunnerInstallations): Exchanges the OAuth code from the GitHub redirect for the GitHub App installations your account can access, each with the connectGrant createRunnerPool takes. Installations this organization already has a pool for are flagged alreadyConnected. A state that does not match this organization answers 400 OAUTH_STATE_INVALID. - [Get a runner pool (with usage)](/docs/runners/getRunnerPool): Returns one runner pool with its RAM tier, disk size, subscription state and live usage: the active and queued runners and the RAM they hold. - [Delete (or drain) a runner pool](/docs/runners/deleteRunnerPool): Deletes a runner pool, or moves it to draining when jobs are still active or queued. The response status says which: deleted or draining. A delete cancels the pool's subscription. A draining pool keeps its subscription until you call delete again once its jobs finish. - [Resize a runner pool's RAM tier](/docs/runners/resizeRunnerPool): Changes a runner pool's RAM tier (poolRamGb) and moves its subscription to the matching price. It cannot go below the RAM its active and queued jobs use. - [List runner jobs for a pool](/docs/runners/listRunnerJobs): Returns a runner pool's jobs, most recently queued first, with repository, runner size, status and timestamps. Pages with limit (default 50) and offset. - [Get runner job metrics and cost comparison](/docs/runners/getRunnerJobMetrics): Returns a runner pool's daily job counts, totals, average queue and queued-to-completed times over 30 days, and the GitHub-hosted cost of its completed jobs. The GitHub estimate covers completed jobs, per started minute of run time. savedCents is that estimate minus the pool's monthly price, never below zero. - Catalog: Public catalog: GPU models, availability, pricing, templates, and locations, plus the OpenAI-compatible model list and public GPU list rates. The list rates need no API key. - [Overview](/docs/catalog/overview): Public catalog: GPU models, availability, pricing, templates, and locations, plus the OpenAI-compatible model list and public GPU list rates. The list rates need no API key. - [List catalog models (OpenAI-compatible)](/docs/catalog/listModels): Returns the model catalog with each model's supported APIs, endpoints, modalities, status and batch eligibility. A model scoped to your organization also carries its token pricing. Free, not balance-gated. Models with status "available" can be used immediately in /v1/chat/completions; "request_access" models are listed for discovery but are not yet invocable. - [GPU availability by model](/docs/catalog/getGpuAvailability): Live availability per GPU model for VMs: total and free GPUs and the largest single allocation placeable right now, served from a short-lived cache. Capacity held exclusively for another organization is left out. Free GPUs released from another organization's reservation to the spot market are counted in spotFreeGpus, not freeGpus. - [GPU model catalog](/docs/catalog/listGpuModels): Returns every GPU model in the catalog with its name, manufacturer, VRAM, architecture, form factor, and interconnect, and whether it is hidden. - [GPU + CPU pricing](/docs/catalog/getPricing): Returns hourly GPU rates per model and tier (with your organization's negotiated rates applied), CPU rates per machine size, and VM disk size bounds. - [Published GPU list rates (no auth)](/docs/catalog/getPublicPricing): List rates for every sellable GPU model (community and verified, per GPU-hour) and online token rates for public models, with no API key required. GPU rates are read from the platform's GPU catalog. Unauthenticated, so it never applies a caller's negotiated overrides: GET /v1/pricing answers with those and keeps its API key. Hidden models, and models missing either tier, are omitted rather than quoted at zero. Served from a short-lived cache, so generatedAt trails the request. - [VM template catalog](/docs/catalog/listVmTemplates): Returns the VM images a VM can be created from, the GPU models each one boots, and which image a VM gets when its create names no template. Each template carries its id (the templateId a create names), name, OS type and version, driver version, whether it needs a GPU, the GPU models it boots, and the GPU models (or CPU VMs) that boot it when a create names no template. An image offered by id only, for validation before it is listed, is not returned. While the control plane has not read the image catalog yet this answers 503 VM_IMAGE_CATALOG_UNAVAILABLE; retry it. - [Locations / regions](/docs/catalog/listLocations): Returns the catalog of locations (regions), each with an id, name, city, country, and region, plus latitude and longitude when they are set. - **Billing and usage** - Billing: Prepaid balance, saved cards, deposits, and auto-recharge: read the balance and recent transactions, save or remove cards, and top up from a saved card. - [Overview](/docs/billing/overview): Prepaid balance, saved cards, deposits, and auto-recharge: read the balance and recent transactions, save or remove cards, and top up from a saved card. - [Org balance + recent transactions](/docs/billing/getBalance): Returns the org's ledger and available balances, accrued and current-month usage, minimum deposit, and up to 25 of its newest transactions. Provider payouts are left out of the transactions. - [Start saving a card (Stripe SetupIntent)](/docs/billing/createBillingSetupIntent): Creates a Stripe SetupIntent to save a card and returns its client secret, the publishable key, and bankCheck, true when the bank must confirm the card. The inline Elements form confirms the client secret. The dashboard's card form sends its device identification and a Turnstile token as optional query parameters, and the card save is judged on them: a device that puts the account on a verification hold gets 403 ACCOUNT_PENDING_VERIFICATION, or 403 VERIFICATION_UNAVAILABLE when the hold could not be recorded. Anything else that looks wrong about the request or the account (no verified device, a missing or failed Turnstile token, a new account, a recent failed card, and more) asks the card's bank to confirm the card with 3D Secure, and bankCheck says so. The operation takes no request body, so a bare POST (the CLI, the MCP tool) still works and its card always needs the bank's confirmation. After stripe.confirmSetup succeeds, call the complete operation to learn whether the card was kept. - [Settle a saved card](/docs/billing/completeBillingSetupIntent): Called right after stripe.confirmSetup succeeds. Settles the card the SetupIntent saved and reports whether it was saved, refused, or held. Settlement runs once: a card tied to fraud, or a second card the bank did not verify when it was asked to, puts the account on hold (held, ACCOUNT_PENDING_VERIFICATION); a card the bank did not verify is removed (refused, CARD_NOT_VERIFIED); any other card becomes the account's default (saved). Calling it again, or after the Stripe webhook settled the card, answers the same outcome and changes nothing. - [Charge the saved card to top up prepaid balance](/docs/billing/createBillingDeposit): Charges the org's default saved card off-session for amountCents and returns the PaymentIntent id and status, which tells the client what to do next. The balance credit is applied by the payment_intent.succeeded webhook (the authoritative money signal), not this response: poll the balance after success. - [Read the org's auto-recharge policy](/docs/billing/getAutoRecharge): Returns the org's auto-recharge settings: whether it is on, the available-balance threshold that triggers a charge, and the amount charged, in cents. - [Set the org's auto-recharge policy](/docs/billing/updateAutoRecharge): Saves the org's auto-recharge settings and returns them. Turning it on needs a default saved card, an amount from $25 to $5,000, and a threshold of 0 or more. - [List saved cards (brand/last4/exp only)](/docs/billing/listBillingPaymentMethods): Returns the org's saved cards (id, brand, last four digits, and expiry month and year) and the id of the default card. Empty when no card is saved. - [Remove a saved card](/docs/billing/deleteBillingPaymentMethod): Detaches a saved card from the org. Removing the default card clears the default, so no charge goes to it. A card that is not the org's returns 404. - [Set a saved card as the default for charges](/docs/billing/setBillingPaymentMethod): Makes a saved card the org's default for off-session charges (deposits and auto-recharge). Another org's card, or one not yet settled as saved, returns 404. - Usage: Current-period usage and cost breakdowns: this month's hours and cost by GPU model and tier, live burn rate and runway, and daily spend over a trailing window. - [Overview](/docs/usage/overview): Current-period usage and cost breakdowns: this month's hours and cost by GPU model and tier, live burn rate and runway, and daily spend over a trailing window. - [Get current-month usage](/docs/usage/getCurrentUsage): Returns the cents and hours used so far this month by GPU model and tier, plus the live burn per hour and the projected hours until the auto-stop floor. - [Get per-day spend over a trailing window](/docs/usage/getDailyUsage): Returns the org's spend per UTC day for the last 1 to 90 days (default 7), split into GPU model, CPU, and Inference segments, plus the prior window's total. This is what the dashboard spend chart shows. Compute cost is split at UTC day boundaries and inference charges land on the day they were billed. Read-only: costs are priced the same way your bill is, never re-priced. - [Get the org's on-demand inference series](/docs/usage/getOrgInferenceSeries): Requests, error rate, latency and time-to-first-token percentiles, and tokens for the org's online on-demand inference over 1h, 6h, 24h, or 7d. Every series shares one time grid, and `models` lists the models with traffic in the range. Without `model` every series is the org total across its on-demand models. Batch and async calls are not counted. Every series also carries a whole-range figure, and the per-point counts add up to it. Data appears within a minute of the org's first on-demand request. These are display metrics, not billing: spend is in the daily usage endpoint. - **Providers** - Provider: For GPU providers: applications, nodes, provisioning tokens, and earnings, plus the one-command node installer, node lifecycle status and offboarding. - [Overview](/docs/provider/overview): For GPU providers: applications, nodes, provisioning tokens, and earnings, plus the one-command node installer, node lifecycle status and offboarding. - [Get the one-command node installer](/docs/provider/providerBootstrap): Returns a self-contained shell installer, run on the host with curl piped to sudo bash, that enrolls the machine as a provider node in the community pool. With a bearer org API key it mints a fresh provisioning token (24 hours, any number of nodes); with ?token=otk_… (e.g. minted from the dashboard) it self-auths and embeds that token instead. name only applies when minting (the token already carries it). The runtime is detected on the box, not chosen here: a bare metal host enrolls as a qemu VM host with VFIO passthrough, and a virtualised guest enrolls as a runc pod host on its native NVIDIA or AMD driver. - [Get the interruptible container installer](/docs/provider/interruptibleBootstrap): Returns a POSIX sh installer that enrolls the container it runs in as an interruptible node. It needs no root, systemd or capabilities to run. Run it inside the container as `curl -fsSL "" | sh` or `wget -qO- "" | sh`. It downloads the interruptible agent for the container's architecture, verifies its sha256, and runs it; the agent enrolls the container and detaches. The token must be an interruptible one (orl provider provisioning-token create --class interruptible). - [Get provider and application status](/docs/provider/getProviderStatus): Returns whether the organization is an approved provider, plus the status and submission time of its provider application when it has one. - [Submit a provider application](/docs/provider/applyForProvider): Submits the organization's application to become a provider, with a required note. Answers 400 if it is already a provider or has an application pending. - [List provisioning tokens (provider only)](/docs/provider/listProvisioningTokens): Returns the organization's provisioning tokens, newest first, with each token's prefix, class, status, expiry and enrollment count. The secret is never shown. A token that can still enroll is always listed. Expired, exhausted and revoked tokens stay listed until 6 days after they expire, then fall out. - [Generate a provisioning token (provider only)](/docs/provider/generateProvisioningToken): Mints a provisioning token for enrolling nodes and returns its secret, shown only once. A persistent token enrolls any number of nodes for 24 hours. An interruptible token (class interruptible) lives 30 days and enrolls at most maxEnrollments nodes. A legacy API key needs clusters:write. - [Revoke a provisioning token](/docs/provider/revokeProvisioningToken): Revokes a provisioning token so it can no longer enroll nodes, and it stays listed as revoked. Nodes the token already enrolled keep running. A legacy API key needs clusters:write. - [List provider nodes with GPUs and location](/docs/provider/listProviderNodes): Returns the organization's nodes, newest first, with hardware, GPUs, location and status. Removed nodes are left out unless includeRemoved is set. A status filter returns only the nodes in that status, removed included. Each node also carries its disk space while its tunnel is up, and a summary of any live reservation on it. - [Set a node's lifecycle status](/docs/provider/updateProviderNode): Sets the lifecycle status of a node your organization owns to one of online, draining, maintenance, or removed, and returns the node as updated. draining and removed also terminate the VMs and pods on the node that are not already stopped; removed ends stopped pods too. A removed node answers 400 NODE_DECOMMISSIONED; re-enroll it to bring it back. - [Offboard an interruptible node](/docs/provider/offboardProviderNode): Moves an interruptible node to draining at once and removes it at the deadline: terminateAt, or now plus noticeSeconds, or now when neither is given (send {}). A deadline only ever moves earlier, so repeating or replaying the call is safe. Use it to pass on a reclaim warning from the market the container runs in, or to take a node out yourself. Persistent nodes are offboarded with `orl provider node update --status removed`. - [Provider node + token stats](/docs/provider/getProviderStats): Returns counts of the organization's nodes (total, online and offline, removed nodes excluded) and of its provisioning tokens that can still enroll a node. - [Get provider earnings and payout history](/docs/provider/getProviderEarnings): Returns the provider's earnings totals, the amount available for payout, breakdowns by node, GPU model and day, a count of active rentals, and recent payouts. - [Terminate spot workloads on a node](/docs/provider/reclaimSpotWorkloads): Terminates spot-tier workloads on a node your organization owns, or on reservations it manages there, and returns the VM ids it terminated and left running. A managing org (managingOrgIds) reaches only the released units of the reservations it manages, while each is active and inside its term. Scoped strictly to workloads bought at the spot tier: a spot instance is sold on the explicit contract that the capacity owner can terminate it at any time. Non-spot workloads on the node are never touched. Every termination is audited and the affected VM records capacity reclaim as its reason, so the customer is not told it was a platform fault. - Reservations: Long-term reservations of specific GPU units on a node for a set of organizations. Reserved capacity is exclusive to its audience and hidden from everyone else. - [Overview](/docs/reservations/overview): Long-term reservations of specific GPU units on a node for a set of organizations. Reserved capacity is exclusive to its audience and hidden from everyone else. - [List a node's reservable GPU units](/docs/reservations/listReservationUnits): Returns the allocatable GPU units on one node, with the unit ids that create, release, and reclaim take, and whether each unit can be reserved. Without this those calls are unusable: they are unit-scoped by design, and nothing else on the API returns a unit id, so the first reservation on a node could not be written at all. Each unit is one IOMMU group and is indivisible: a paired unit sells its GPUs together. Sellable units of a reservable class (vfio-group for VMs, device-request for pods) are reservable, and the response says which, and why not, rather than leaving the caller to infer it from a rejection. Readable by a platform operator, the org that owns the node, and any org managing a reservation on it. Anyone else gets 404, so the fleet stays unenumerable. - [List a node's reservations](/docs/reservations/listNodeReservations): Returns every reservation on one node, each with its reserved unit ids, the units released to spot, its audience, managing orgs, term and status. Readable by a platform operator, the org that owns the node, and any org managing a reservation on it that is active and inside its term. Anyone else gets 404. - [Reserve GPU units for orgs (platform admin)](/docs/reservations/createNodeReservation): Grants exclusive placement rights over specific GPU units for a term. Reserved units disappear from every other org's catalog and cannot be placed on by them. Overlap with an existing reservation is rejected atomically. - [Get one node reservation](/docs/reservations/getNodeReservation): Returns one reservation on a node: its reserved unit ids, the units currently released to spot, its audience, managing orgs, term and status. Readable by the same orgs as the list: a platform operator, the org that owns the node, and any org managing a reservation on it that is active and inside its term. - [Amend a reservation (platform admin)](/docs/reservations/updateNodeReservation): Changes a reservation's term end, audience, or managing orgs, or re-pins it to the node's current topology, and returns the updated reservation. Re-pinning (reaffirm) also reactivates a reservation in needs-review. The start of the term is fixed at create. - [End a reservation now (platform admin)](/docs/reservations/cancelNodeReservation): Ends a reservation now and returns its GPU units to the public pool immediately. Workloads the tenant is already running keep running. The node's owning org terminates those workloads when it needs the capacity. - [Release GPU units to the spot market](/docs/reservations/releaseReservationUnits): Puts the named GPU units of a reservation on the spot market. They stay part of the reservation, and anyone may buy them, but only at tier=spot. This is the double-sell operation: spot-only sale is what makes them terminable by the capacity owner at any time. An ordinary community or verified request will never be placed on them, so released capacity can always be reclaimed. Callable by a platform operator, the org that owns the node, or an org in the reservation's managingOrgIds while the reservation is active and inside its term. - [Reclaim released GPU units from spot](/docs/reservations/reclaimReservationUnits): Returns the named units to the reservation's exclusive use and terminates the spot workloads running on them, in one call. Everything evicted was bought at the spot tier, whose terms are exactly that the capacity owner can reclaim without notice. Pass dryRun to see what would be terminated without changing anything. Callable by a platform operator, the org that owns the node, or an org in the reservation's managingOrgIds while the reservation is active and inside its term. - [List GPU capacity reserved to this org](/docs/reservations/listOrgReservations): Returns the GPU reservations this organization is in the audience of or manages, with node ids, reserved units and terms. Expired and cancelled ones are left out; pending and needs-review reservations are included. With reservations not enabled for the org, the list is empty. - **Inference** - Chat Completions: Generate a model response for a conversation, OpenAI-compatible: the core online inference family, with `orl chat` as its friendly front. Served by the inference endpoint (inference.openrelay.inc). - [Overview](/docs/chat-completions/overview): Generate a model response for a conversation, OpenAI-compatible: the core online inference family, with `orl chat` as its friendly front. Served by the inference endpoint (inference.openrelay.inc). - [Generate a chat completion](/docs/chat-completions/createChatCompletion): Generates a model response for a conversation in the OpenAI Chat Completions shape, from any model whose supported APIs include chat completions. The hand-written `orl chat` is the friendlier front for this operation: model name completion, a readable transcript, and streaming output rendered to the terminal. Requires a funded balance. Set "stream" to true to receive the completion incrementally as Server-Sent Events, ending with a `data: [DONE]` sentinel. The request body is capped at 10 MiB. - Messages: Generate a model response in the Anthropic Messages shape for models that support it: a drop-in for the Anthropic SDKs and Claude Code. Served by the inference endpoint (inference.openrelay.inc). - [Overview](/docs/messages/overview): Generate a model response in the Anthropic Messages shape for models that support it: a drop-in for the Anthropic SDKs and Claude Code. Served by the inference endpoint (inference.openrelay.inc). - [Generate a message (Anthropic-compatible)](/docs/messages/createMessage): Generates a model response in the Anthropic Messages shape, a drop-in for the Anthropic SDKs and Claude Code on models that support messages. Messages is a second online family alongside Chat Completions; check a model's supported_apis on GET /v1/models before calling it rather than assuming. Requires a funded balance. Send the API key as x-api-key (the Anthropic SDK convention) or as Authorization Bearer; either is accepted. Set "stream" to true for event-typed Server-Sent Events (message_start, content_block_delta, message_stop, and so on). The request body is capped at 10 MiB. - Responses: Create a model response in the OpenAI Responses shape for models that support it, stateless: a drop-in for the OpenAI SDKs and Codex. Served by the inference endpoint (inference.openrelay.inc). - [Overview](/docs/responses/overview): Create a model response in the OpenAI Responses shape for models that support it, stateless: a drop-in for the OpenAI SDKs and Codex. Served by the inference endpoint (inference.openrelay.inc). - [Create a model response (OpenAI Responses)](/docs/responses/createResponse): Generates a model response in the OpenAI Responses shape for the OpenAI SDKs and Codex. Check a model's supported_apis on GET /v1/models first. Stateless: the gateway sends store false and refuses previous_response_id, conversation, background, stored prompts, item references and file inputs (400 unsupported_parameter). Send the full conversation in input on every turn. Client-executed tools (function, custom, namespace, client tool_search) are forwarded; hosted web_search is removed and every other hosted tool is refused (400 unsupported_tool_type). Requires a funded balance. Set "stream" to true for event-typed Server-Sent Events (response.created, response.output_text.delta, response.completed, and so on); a stream carries no [DONE] line. The request body is capped at 10 MiB. - Embeddings: Turn text into vectors for search, retrieval, clustering and deduplication, OpenAI-compatible, from the multilingual openrelay/bge-m3. Served by the inference endpoint (inference.openrelay.inc). - [Overview](/docs/embeddings/overview): Turn text into vectors for search, retrieval, clustering and deduplication, OpenAI-compatible, from the multilingual openrelay/bge-m3. Served by the inference endpoint (inference.openrelay.inc). - [Turn text into vectors](/docs/embeddings/createEmbedding): Turns text into vectors for search, retrieval, clustering and deduplication in the OpenAI embeddings shape, served by openrelay/bge-m3 today. openrelay/bge-m3 is multilingual and returns 1024-dimension dense vectors; check a model's supported_apis on GET /v1/models for embeddings. Requires a funded balance. No output rate: billed only on usage.prompt_tokens. stream is dropped. Limits: 8192 tokens per input item, 1024 items per request, 10 MiB request body. Also runs through the Batch API at endpoint /v1/embeddings, at the batch rate. - Decisions: Answer typed questions about an input (a choice, a score or a yes/no) with probabilities and no generated text, online only and billed per input token. Served by the inference endpoint (inference.openrelay.inc). - [Overview](/docs/decisions/overview): Answer typed questions about an input (a choice, a score or a yes/no) with probabilities and no generated text, online only and billed per input token. Served by the inference endpoint (inference.openrelay.inc). - [Answer typed questions about an input](/docs/decisions/createDecision): Answers typed questions about a text or JSON input: a choice, a score or a yes/no, with probabilities. Bills input tokens only. Each answer carries its probabilities and a confidence; nothing is generated. Served by openrelay/qwen3.6-35b-decide, which reads the probability of each option's letter at one token; check a model's supported_apis on GET /v1/models for decisions. Online only: the Batch API does not accept /v1/decisions. Requires a funded balance. Any field this schema does not list is rejected with 400 unknown_field. stream is dropped: a decision is always a single JSON response. Limits: 1 to 64 questions per request, 10 MiB request body. - Batches: Batch inference jobs: submit a JSONL file of requests, poll status, and download results. A batch has 24 hours to finish; standard batches bill at a discount. Served by the inference endpoint (inference.openrelay.inc). Batch access is enabled per organization; without it these endpoints return 404. - [Overview](/docs/batches/overview): Batch inference jobs: submit a JSONL file of requests, poll status, and download results. A batch has 24 hours to finish; standard batches bill at a discount. Served by the inference endpoint (inference.openrelay.inc). Batch access is enabled per organization; without it these endpoints return 404. - [Get the Batch API's current limits](/docs/batches/getBatchLimits): Returns the limits a batch submission runs against today: the per-batch minimum fee and the metadata caps that create and update both enforce. The minimum fee is 0 when none is configured. orl batches create reads this instead of assuming fixed constants, so a limit change on the server reaches it without a CLI release. - [List batches, newest first](/docs/batches/listBatches): Returns one page of your organization's batches, newest first, in the OpenAI list envelope. Page with limit (1 to 100, default 20) and after. - [Create a batch inference job](/docs/batches/createBatch): Submit a batch from an uploaded input file (input_file_id) or inline request objects (requests); provide exactly one of the two. Every record is checked when the batch is created. A record that breaks a rule is rejected and the rest are accepted: the answer counts the rejections in rejected_count, lists the first 100 in rejected_records, and rejected_records_file_id names a file holding all of them. A rejected record never runs, is never billed, counts as failed in request_counts, and is also listed in the batch's error file. When no record can run, the answer is 400 no_valid_records with the same fields under error, and nothing is created. A record carrying openrelay.workflow from an organization without workflow records, or priority expedited where it is turned off, refuses the whole submission with 409 feature_disabled, before any record is checked: a workflow record anywhere in the input wins over input_too_large, a read failure after that record, no_valid_records and every record rejection. A batch has 24 hours from submission (expires_at) to finish: records not done by then come back with the batch_expired error and the batch ends expired. Standard batches bill at the discounted batch rate. Poll it with get, then download results via the output and error file ids. - [Get a batch by id](/docs/batches/getBatch): Returns one batch with its status, priority and request counts, plus its usage once any is recorded and the output and error file ids once written. Poll it to track progress. - [Cancel a running batch](/docs/batches/cancelBatch): Cancels a batch that has not finished. In-flight records finish, unstarted ones are skipped, and partial results are written; completed work is billed. The batch settles to cancelled within a few minutes. A batch already in a terminal state returns 409. - [Replace a batch's metadata](/docs/batches/updateBatchMetadata): Replace the batch's metadata map with the one you send; the whole map is overwritten, so include every pair you want to keep. Send an empty object to clear it. Metadata is opaque customer data, not lifecycle state, so it is editable at any time including after the batch is terminal. Returns the updated batch. - [Workflow record callback](/docs/batches/batchWorkflowEvent): The events OpenRelay POSTs to a workflow record's callback, and the answers it accepts: the next requests to run, a pause, or the outcome. A workflow record is a batch record whose body carries openrelay.workflow. OpenRelay POSTs an event to its workflow.url, runs the requests the answer returns, and POSTs their results as the next event, until the callback answers completed or failed. The URL is the record's own, not a registered webhook. The request carries the record's workflow.authorization value verbatim as its Authorization header, and Content-Type application/json; it is not signed. Each event is delivered up to four times, with pauses of 1, 2 and 3 seconds between attempts and each attempt bounded by callback_timeout_ms. A timeout, a connection error, a 429 or a 5xx answer is retried, and so is a redirect, which is never followed. When all four attempts fail, the record fails with workflow_callback_unreachable and the rest of the batch keeps running. Any other non-2xx answer fails the record with workflow_callback_rejected. - Async: Async jobs: requests delivered within 1 hour (20% off the Sync price) or within 24 hours (50% off), with their results paged while the job runs. Served by the inference endpoint (inference.openrelay.inc). These endpoints return 404 for an organization Async is not turned on for. - [Overview](/docs/async/overview): Async jobs: requests delivered within 1 hour (20% off the Sync price) or within 24 hours (50% off), with their results paged while the job runs. Served by the inference endpoint (inference.openrelay.inc). These endpoints return 404 for an organization Async is not turned on for. - [List async jobs](/docs/async/listAsyncJobs): Your organization's async jobs, newest first, a page at a time. Add status to list only the jobs with that status. A request with status reads a bounded number of jobs, so a page can hold fewer than limit, or none, with has_more true. Page on with after set to its last_id, which is then the last job the request read and may not be on the page. - [Create an async job](/docs/async/createAsyncJob): Submit requests as a job, for delivery within 1 hour (20% off the Sync price) or within 24 hours (50% off Sync). Send them inline (requests, up to 1,000) or as an uploaded input file (input_file_id); provide exactly one of the two. Each line names its url (/v1/chat/completions, /v1/messages or /v1/embeddings) and the body that endpoint takes, and a job may mix models and endpoints. Every line is checked when the job is created, with the Batch API's record rules: a line that breaks one is rejected and the rest are accepted. The answer counts the rejections in rejected_count, lists the first 100 in rejected_records, and rejected_records_file_id names a file holding all of them; a rejected line never runs, is never billed, and counts as failed. When no line can run, the answer is 400 no_valid_records and nothing is created. A request not started before deadline_at expires at no charge; one that started finishes and is charged. A 1-hour job holds up to 10,000 requests and a 24-hour job up to 50,000. - [Get an async job](/docs/async/getAsyncJob): The job, its status, request counts, token usage and the cost billed so far, readable while it runs and after it ends. - [Cancel an async job](/docs/async/cancelAsyncJob): Stop starting the job's requests. Requests already running finish and are charged; the rest end cancelled at no charge. A job that has not started ends cancelled at once; a running one answers cancelling until its in-flight requests finish. - [Read an async job's results](/docs/async/listAsyncJobResults): Finished results so far, a page at a time, readable while the job runs: results arrive a group of requests at a time, as each group finishes. Groups show in order. A page reads at most 32 MB of results plus the one result that crosses that budget (a result is never split), and can end with none and has_more true; poll on with its next_cursor. has_more stays true while the job is live; poll with the same after. A cursor pages only the job and the status filter it was read under. - Files: Input and result files for the Batch API (JSONL, up to 200 MB and 50,000 records). Served by the inference endpoint (inference.openrelay.inc). - [Overview](/docs/files/overview): Input and result files for the Batch API (JSONL, up to 200 MB and 50,000 records). Served by the inference endpoint (inference.openrelay.inc). - [Upload a JSONL input file for the Batch API](/docs/files/uploadFile): Uploads a JSONL file of batch requests and returns its file object: the file id, size in bytes, purpose, and an expiry time 30 days after the upload. Send multipart/form-data with a `file` part (the JSONL bytes, up to 200 MB and 50,000 records) and a `purpose` field set to `batch`. Each line is one request object with the BatchRequestRecord shape (including the optional `openrelay.tool_config` extension on chat bodies); the returned file id goes in `input_file_id` when creating a batch. The upload checks only that the file is JSONL. Its records are checked when a batch is created from it, which rejects the invalid ones and runs the rest. - [Mint a signed URL for a direct JSONL upload](/docs/files/presignFileUpload): For browser clients that upload the file straight to storage instead of through the API: returns a file id and a signed PUT URL (valid 15 minutes). After the PUT succeeds, register the file with the complete call. CLI and SDK clients should use the plain upload instead. - [Register a file after its presigned upload](/docs/files/completeFileUpload): Checks an object uploaded to a presigned URL as the plain upload checks a file, and registers it as a batch input file to use in a batch. The checks are its size, its record count, and that it is JSONL; an object they refuse is deleted. Idempotent; completing an already-registered id returns the existing file. - [Get a file's metadata](/docs/files/getFile): Returns a file's metadata (id, size in bytes, purpose, and creation and expiry times) for a batch input, result or rejections file, without its content. The filename is assigned by the server, not the name you uploaded. - [Download a file's raw JSONL content](/docs/files/getFileContent): Streams a file's raw bytes as application/octet-stream, for a batch input file you uploaded, a result file, or a batch create's rejections file. Use it to fetch batch results: the batch object's output_file_id (successful records) and error_file_id (failed records) each point at a JSONL file of `{custom_id, response | error}` lines. A create's rejected_records_file_id points at a JSONL file of BatchRecordRejection lines, one per record the create rejected. - Dedicated Endpoints: Models served on capacity reserved for your organization. Call one through the inference API at inference.openrelay.inc with its modelId. Read it, its lifecycle events, and its monitoring series here. Dedicated endpoints are set up with our team; without dedicated inference the list is empty and the rest return 404. - [Overview](/docs/dedicated-endpoints/overview): Models served on capacity reserved for your organization. Call one through the inference API at inference.openrelay.inc with its modelId. Read it, its lifecycle events, and its monitoring series here. Dedicated endpoints are set up with our team; without dedicated inference the list is empty and the rest return 404. - [List your dedicated endpoints](/docs/dedicated-endpoints/listDedicatedEndpoints): Returns one page of your organization's dedicated inference endpoints, with each one's model id, status, enabled flag, and per-token pricing. - [Get a dedicated endpoint](/docs/dedicated-endpoints/getDedicatedEndpoint): Returns one of your dedicated endpoints: the model id to send in inference requests, its status, whether it is enabled, and its per-token pricing. - [List a dedicated endpoint's events](/docs/dedicated-endpoints/listDedicatedEndpointEvents): Returns one page of a dedicated endpoint's lifecycle events (created, updated, enabled or disabled), each with its action, reason and time. - [Get a dedicated endpoint's monitoring series](/docs/dedicated-endpoints/getDedicatedEndpointSeries): Requests, error rate, latency and time-to-first-token percentiles, and tokens for a dedicated endpoint over 1h, 6h, 24h, or 7d, on one time grid. Every series also carries a whole-range figure. This is what the dashboard's Monitoring tab shows. Data appears within a minute of the endpoint serving its first request. The example cuts each `values` array to its first three points; a `1h` response has 241. - [Get a dedicated endpoint's usage for a month](/docs/dedicated-endpoints/getDedicatedEndpointUsage): A dedicated endpoint's GPU-hours and their cost for one calendar month (UTC), read from the usage records its monthly bill is settled from. It reports the GPU-hours the endpoint's replicas were hosted, the GPU-hours of those that were billed (a replica is billed while it answers requests), and their cost in cents. The current month runs to now. A deleted endpoint's usage stays readable for the months it has records in. It is a billing read: it needs the usage permission, as the organization's usage routes do. Requests to the endpoint are not charged per token, so this is the endpoint's whole cost. It is rounded once for the endpoint, and the bill once for the organization, so the two can differ by under a cent per endpoint. - Model Imports: The credentials that imports of your own model weights read private Hugging Face, S3 and Google Cloud Storage sources with, for dedicated endpoints. Without dedicated inference these return 404. - [Overview](/docs/model-imports/overview): The credentials that imports of your own model weights read private Hugging Face, S3 and Google Cloud Storage sources with, for dedicated endpoints. Without dedicated inference these return 404. - [List your organization's model imports](/docs/model-imports/listModelImports): Every model import of your organization, running or ended, each with its status, progress, source and the artifact it produces. Read the artifact for its facts and the hardware it fits. Imports are listed by id, not by when they started. - [Import model weights](/docs/model-imports/createModelImport): Copies a model's weights from Hugging Face, S3 or Google Cloud Storage into OpenRelay and checks them, to serve on a dedicated endpoint. Name a credential for a private source. attestRights must be true: you attest you have the rights to the weights, under the license you name. The import runs in the background; poll it until it succeeds or fails. Your organization runs a limited number of imports at once (409 IMPORT_LIMIT names it), and a create passes the same balance and account checks as a VM create. - [Get a model import](/docs/model-imports/getModelImport): One of your organization's model imports: its status, progress, the revision it pinned, and why it failed, with the artifact it produces. Another organization's import reads as not found. - [List your organization's model artifacts](/docs/model-imports/listModelImportArtifacts): The weights your organization's imports produced: each artifact's status, facts and the hardware it was found to fit. A ready artifact is what a dedicated endpoint of the import kind serves. Artifacts are listed by id. - [Get a model artifact](/docs/model-imports/getModelImportArtifact): One of your organization's model artifacts: its architecture, dtype, quantization, size, and the hardware it was found to fit or refused. Another organization's artifact reads as not found. - [List an organization's model import credentials](/docs/model-imports/listModelImportCredentials): The credentials the organization's model imports read their sources with: name, type and a hint to tell two apart, never a secret. An s3_role also shows what the customer's role must trust, and a gcs_federation what the customer's pool must admit. - [Create a model import credential](/docs/model-imports/createModelImportCredential): Saves a credential your model imports read a private source with: a Hugging Face token, an S3 role or key pair, or a Google Cloud credential. A Google Cloud credential is a workload identity pool provider or a service account key. A secret is sealed with a key only the organization's import sessions can use, and is never returned. An s3_role answers with the external id issued for it and the role its trust policy must admit; a gcs_federation with the subject its pool must admit. A field the type does not take, or a malformed secret, answers 400. An organization holds at most 50, and a role or pool provider another organization registered answers 409 CREDENTIAL_CLAIMED. - [Delete a model import credential](/docs/model-imports/deleteModelImportCredential): Removes a model import credential and its sealed secret. A credential that a running import reads its source with cannot be deleted yet. It answers 204. One a running import reads answers 409 CREDENTIAL_IN_USE; delete it once that import has finished. - CLI: The orl command line - [Overview](/docs/cli): orl is the OpenRelay command-line tool. Log in with an API key, deploy GPU VMs, open SSH sessions, and drive the whole API from your terminal or an agent. - [Quickstart](/docs/cli/quickstart): Sign up, log in, and get a GPU VM running with the orl CLI in four commands, then connect to it, watch its spend, and stop it when you are done. - [Authentication](/docs/cli/authentication): How orl signs you in (a browser device login, or an API key for scripts and CI), picks your organization, and which setting wins when several are set. - [Workflows](/docs/cli/workflows): The signature orl commands for deploying and connecting to GPU VMs and watching spend, plus how to script orl in CI with JSON errors and exit codes. - [Kubernetes](/docs/cli/kubernetes): orl kube manages Managed Kubernetes clusters, node pools and nodes, and writes kubeconfigs that authenticate kubectl with your OpenRelay API key. - [Agents (MCP)](/docs/cli/agents): Run orl from an agent, as commands or as MCP tools, with JSON errors and no prompts. Add it to Claude and choose which tools to expose. - [Upgrade & shell setup](/docs/cli/upgrade): Keep the orl CLI current with orl upgrade or Homebrew, and add shell completions for bash, zsh, and fish, plus man pages. - [Changelog](/docs/cli/changelog): What changed in each orl release, with links to the generated GitHub release notes and the exact command to install a specific version. - Commands - [orl account invite accept](/docs/cli/commands/account-invite-accept): Accept a pending org invite (session only) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl account invite decline](/docs/cli/commands/account-invite-decline): Decline a pending org invite (session only) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl account invite list](/docs/cli/commands/account-invite-list): List pending invites to the signed-in user (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl account profile update](/docs/cli/commands/account-profile-update): Update the caller's profile (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl account whoami](/docs/cli/commands/account-whoami): The current principal and the org it acts in, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl api](/docs/cli/commands/api): Call a control-plane or inference-host path with the current credentials, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl api-keys create](/docs/cli/commands/api-keys-create): Create an API key (returns the plaintext once) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl api-keys delete](/docs/cli/commands/api-keys-delete): Revoke an API key (needs `orl auth login`) (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl api-keys list](/docs/cli/commands/api-keys-list): List an org's API keys (no secrets), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl async job cancel](/docs/cli/commands/async-job-cancel): Cancel an async job (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl async job create](/docs/cli/commands/async-job-create): Create an async job, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl async job get](/docs/cli/commands/async-job-get): Get an async job, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl async job list](/docs/cli/commands/async-job-list): List async jobs, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl async job-result list](/docs/cli/commands/async-job-result-list): Read an async job's results, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl audit-events list](/docs/cli/commands/audit-events-list): List the org's access audit log, newest first, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl auth login](/docs/cli/commands/auth-login): Sign in (device login on a terminal; --with-token for an API key), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl auth logout](/docs/cli/commands/auth-logout): Remove stored credentials for the current API host, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl auth status](/docs/cli/commands/auth-status): Show where credentials come from and whether they work, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl authz check-access](/docs/cli/commands/authz-check-access): Check whether a principal may perform an action, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl authz effective get](/docs/cli/commands/authz-effective-get): Show what a principal can do right now, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl authz key-setting get](/docs/cli/commands/authz-key-setting-get): Get the organization's key settings, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl authz key-setting update](/docs/cli/commands/authz-key-setting-update): Set the maximum key lifetime (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl authz permission-catalog get](/docs/cli/commands/authz-permission-catalog-get): Get the permission catalog, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl authz policy create](/docs/cli/commands/authz-policy-create): Create a guardrail or a custom Cedar policy (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl authz policy delete](/docs/cli/commands/authz-policy-delete): Delete a policy (needs `orl auth login`) (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl authz policy get](/docs/cli/commands/authz-policy-get): Get a guardrail or custom policy, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl authz policy list](/docs/cli/commands/authz-policy-list): List the org's guardrails and custom policies, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl authz policy update](/docs/cli/commands/authz-policy-update): Update, enable or disable a policy (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl authz policy validate](/docs/cli/commands/authz-policy-validate): Validate a policy without saving it (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl balance](/docs/cli/commands/balance): Balance and current-month spend at a glance, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl batches cancel](/docs/cli/commands/batches-cancel): Cancel a running batch (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl batches create](/docs/cli/commands/batches-create): Create a batch inference job, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl batches get](/docs/cli/commands/batches-get): Get a batch by id, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl batches limit get](/docs/cli/commands/batches-limit-get): Get the Batch API's current limits, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl batches list](/docs/cli/commands/batches-list): List batches, newest first, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl batches metadata update](/docs/cli/commands/batches-metadata-update): Replace a batch's metadata, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl batches results](/docs/cli/commands/batches-results): Download a finished batch's output and error records, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl batches wait](/docs/cli/commands/batches-wait): Wait for a batch to finish, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl billing auto-recharge get](/docs/cli/commands/billing-auto-recharge-get): Read the org's auto-recharge policy, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl billing auto-recharge update](/docs/cli/commands/billing-auto-recharge-update): Set the org's auto-recharge policy (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl billing balance get](/docs/cli/commands/billing-balance-get): Org balance + recent transactions, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl billing deposit create](/docs/cli/commands/billing-deposit-create): Charge the saved card to top up prepaid balance (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl billing payment-method delete](/docs/cli/commands/billing-payment-method-delete): Remove a saved card (needs `orl auth login`) (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl billing payment-method list](/docs/cli/commands/billing-payment-method-list): List saved cards (brand/last4/exp only), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl billing payment-method set](/docs/cli/commands/billing-payment-method-set): Set a saved card as the default for charges (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl billing setup-intent create](/docs/cli/commands/billing-setup-intent-create): Start saving a card (Stripe SetupIntent) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl catalog gpu-availability get](/docs/cli/commands/catalog-gpu-availability-get): GPU availability by model, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl catalog gpu-model list](/docs/cli/commands/catalog-gpu-model-list): GPU model catalog, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl catalog location list](/docs/cli/commands/catalog-location-list): Locations / regions, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl catalog model list](/docs/cli/commands/catalog-model-list): List catalog models (OpenAI-compatible), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl catalog pricing get](/docs/cli/commands/catalog-pricing-get): GPU + CPU pricing, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl catalog vm-template list](/docs/cli/commands/catalog-vm-template-list): VM template catalog, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl chat](/docs/cli/commands/chat): Send one chat completion to a model and print the answer, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl chat-completions create](/docs/cli/commands/chat-completions-create): Generate a chat completion, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl decisions create](/docs/cli/commands/decisions-create): Answer typed questions about an input, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl dedicated-endpoints event list](/docs/cli/commands/dedicated-endpoints-event-list): List a dedicated endpoint's events, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl dedicated-endpoints get](/docs/cli/commands/dedicated-endpoints-get): Get a dedicated endpoint, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl dedicated-endpoints list](/docs/cli/commands/dedicated-endpoints-list): List your dedicated endpoints, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl dedicated-endpoints series get](/docs/cli/commands/dedicated-endpoints-series-get): Get a dedicated endpoint's monitoring series, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl dedicated-endpoints usage get](/docs/cli/commands/dedicated-endpoints-usage-get): Get a dedicated endpoint's usage for a month, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl deploy](/docs/cli/commands/deploy): Create a GPU VM with a guided or scripted flow, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl doctor](/docs/cli/commands/doctor): Check the install, credentials, both API hosts, the key's access, and the balance, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl embeddings create](/docs/cli/commands/embeddings-create): Turn text into vectors, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl files content get](/docs/cli/commands/files-content-get): Download a file's raw JSONL content, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl files get](/docs/cli/commands/files-get): Get a file's metadata, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl files upload](/docs/cli/commands/files-upload): Upload a JSONL input file for the Batch API, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl imports artifact get](/docs/cli/commands/imports-artifact-get): Get a model artifact, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl imports artifact list](/docs/cli/commands/imports-artifact-list): List your organization's model artifacts, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl imports create](/docs/cli/commands/imports-create): Import model weights, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl imports credential create](/docs/cli/commands/imports-credential-create): Create a model import credential, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl imports credential delete](/docs/cli/commands/imports-credential-delete): Delete a model import credential (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl imports credential list](/docs/cli/commands/imports-credential-list): List an organization's model import credentials, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl imports get](/docs/cli/commands/imports-get): Get a model import, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl imports list](/docs/cli/commands/imports-list): List your organization's model imports, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl init](/docs/cli/commands/init): Set up a project for OpenRelay: env template, and with --agent an AGENTS.md block and MCP config, part of the orl CLI for OpenRelay GPU VMs, pods, batches and… - [orl kube cluster create](/docs/cli/commands/kube-cluster-create): Create a Kubernetes cluster, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl kube cluster delete](/docs/cli/commands/kube-cluster-delete): Delete a Kubernetes cluster (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl kube cluster get](/docs/cli/commands/kube-cluster-get): Get a Kubernetes cluster, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl kube cluster list](/docs/cli/commands/kube-cluster-list): List Kubernetes clusters, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl kube kubeconfig](/docs/cli/commands/kube-kubeconfig): Write a kubeconfig that authenticates with orl kube token, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl kube node delete](/docs/cli/commands/kube-node-delete): Delete a node (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl kube node get](/docs/cli/commands/kube-node-get): Get a node, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl kube node list](/docs/cli/commands/kube-node-list): List a cluster's nodes, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl kube node-pool create](/docs/cli/commands/kube-node-pool-create): Add a node pool to a cluster, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl kube node-pool delete](/docs/cli/commands/kube-node-pool-delete): Delete a node pool (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl kube node-pool get](/docs/cli/commands/kube-node-pool-get): Get a node pool, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl kube node-pool list](/docs/cli/commands/kube-node-pool-list): List a cluster's node pools, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl kube node-pool scale](/docs/cli/commands/kube-node-pool-scale): Scale a node pool (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl kube token](/docs/cli/commands/kube-token): Print a cluster token as a kubectl ExecCredential, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl mcp serve](/docs/cli/commands/mcp-serve): Serve OpenRelay API tools over MCP stdio, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl messages create](/docs/cli/commands/messages-create): Generate a message (Anthropic-compatible), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl models](/docs/cli/commands/models): List the models you can call, with their prices and whether they can run as a batch, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl organizations create](/docs/cli/commands/organizations-create): Create an organization (caller becomes owner) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl organizations delete](/docs/cli/commands/organizations-delete): Soft-delete an organization (owner only) (needs `orl auth login`) (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods,… - [orl organizations get](/docs/cli/commands/organizations-get): Get an organization, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl organizations invite delete](/docs/cli/commands/organizations-invite-delete): Revoke a pending invite (owner/admin) (needs `orl auth login`) (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods,… - [orl organizations invite list](/docs/cli/commands/organizations-invite-list): List an org's pending invites (owner/admin) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl organizations member add](/docs/cli/commands/organizations-member-add): Invite a member by email (owner/admin) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl organizations member delete](/docs/cli/commands/organizations-member-delete): Remove a member (owner/admin) (needs `orl auth login`) (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and… - [orl organizations member leave](/docs/cli/commands/organizations-member-leave): Leave an organization (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl organizations member list](/docs/cli/commands/organizations-member-list): List an org's members, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl organizations member-role update](/docs/cli/commands/organizations-member-role-update): Update a member's role (owner/admin) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl organizations switch](/docs/cli/commands/organizations-switch): Set the active organization for future commands, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl organizations update](/docs/cli/commands/organizations-update): Update an organization (owner/admin) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl pods capacity list](/docs/cli/commands/pods-capacity-list): Live pod GPU capacity and on-demand rates, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl pods create](/docs/cli/commands/pods-create): Create a pod, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl pods list](/docs/cli/commands/pods-list): List an org's pods (cursor-paginated), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl pods log get](/docs/cli/commands/pods-log-get): Get a pod's container log tail, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl provider apply](/docs/cli/commands/provider-apply): Submit a provider application, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl provider earnings get](/docs/cli/commands/provider-earnings-get): Get provider earnings and payout history, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl provider node list](/docs/cli/commands/provider-node-list): List provider nodes with GPUs and location, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl provider node offboard](/docs/cli/commands/provider-node-offboard): Offboard an interruptible node (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl provider node update](/docs/cli/commands/provider-node-update): Set a node's lifecycle status (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl provider provisioning-token create](/docs/cli/commands/provider-provisioning-token-create): Generate a provisioning token (provider only), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl provider provisioning-token delete](/docs/cli/commands/provider-provisioning-token-delete): Revoke a provisioning token (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl provider provisioning-token list](/docs/cli/commands/provider-provisioning-token-list): List provisioning tokens (provider only), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl provider spot-workload reclaim](/docs/cli/commands/provider-spot-workload-reclaim): Terminate spot workloads on a node (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl provider stats get](/docs/cli/commands/provider-stats-get): Provider node + token stats, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl provider status get](/docs/cli/commands/provider-status-get): Get provider and application status, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl registry-credentials create](/docs/cli/commands/registry-credentials-create): Create a registry credential, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl registry-credentials delete](/docs/cli/commands/registry-credentials-delete): Delete a registry credential (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl registry-credentials list](/docs/cli/commands/registry-credentials-list): List an org's registry credentials (no secrets), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl reservations list](/docs/cli/commands/reservations-list): List GPU capacity reserved to this org, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl reservations node get](/docs/cli/commands/reservations-node-get): Get one node reservation, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl reservations node list](/docs/cli/commands/reservations-node-list): List a node's reservations, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl reservations unit list](/docs/cli/commands/reservations-unit-list): List a node's reservable GPU units, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl reservations unit reclaim](/docs/cli/commands/reservations-unit-reclaim): Reclaim released GPU units from spot (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl reservations unit release](/docs/cli/commands/reservations-unit-release): Release GPU units to the spot market (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl responses create](/docs/cli/commands/responses-create): Create a model response (OpenAI Responses), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl roles create](/docs/cli/commands/roles-create): Create a custom role (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl roles delete](/docs/cli/commands/roles-delete): Delete a custom role (needs `orl auth login`) (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl roles get](/docs/cli/commands/roles-get): Get a role, with its generated Cedar, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl roles list](/docs/cli/commands/roles-list): List built-in and custom roles, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl roles update](/docs/cli/commands/roles-update): Update a custom role (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl runners connect-url get](/docs/cli/commands/runners-connect-url-get): Get the GitHub connect URL, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl runners installation list](/docs/cli/commands/runners-installation-list): List GitHub installations you can connect, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl runners job list](/docs/cli/commands/runners-job-list): List runner jobs for a pool, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl runners job-metrics get](/docs/cli/commands/runners-job-metrics-get): Get runner job metrics and cost comparison, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl runners pool create](/docs/cli/commands/runners-pool-create): Create a runner pool, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl runners pool delete](/docs/cli/commands/runners-pool-delete): Delete (or drain) a runner pool (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl runners pool get](/docs/cli/commands/runners-pool-get): Get a runner pool (with usage), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl runners pool list](/docs/cli/commands/runners-pool-list): List runner pools (with active/queued usage), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl runners pool resize](/docs/cli/commands/runners-pool-resize): Resize a runner pool's RAM tier, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl schema](/docs/cli/commands/schema): Print JSON Schemas for commands: arguments, flags and output, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl snapshots delete](/docs/cli/commands/snapshots-delete): Soft-delete a snapshot (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl snapshots list](/docs/cli/commands/snapshots-list): List VM snapshots, optionally by source VM, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl ssh-keys create](/docs/cli/commands/ssh-keys-create): Add an SSH public key (computes fingerprint), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl ssh-keys delete](/docs/cli/commands/ssh-keys-delete): Delete an SSH public key (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl ssh-keys list](/docs/cli/commands/ssh-keys-list): List an org's SSH public keys, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl transfers accept](/docs/cli/commands/transfers-accept): Accept an incoming transfer (owner/admin) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl transfers cancel](/docs/cli/commands/transfers-cancel): Cancel a pending outgoing transfer (source org) (needs `orl auth login`) (asks for confirmation; --yes to skip). - [orl transfers create](/docs/cli/commands/transfers-create): Start a resource transfer (owner/admin) (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl transfers incoming list](/docs/cli/commands/transfers-incoming-list): Pending incoming transfers (enriched), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl transfers outgoing list](/docs/cli/commands/transfers-outgoing-list): Outgoing transfers, all statuses (enriched), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl transfers pending-count get](/docs/cli/commands/transfers-pending-count-get): Count of pending incoming transfers, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl transfers reject](/docs/cli/commands/transfers-reject): Reject an incoming transfer (owner/admin) (needs `orl auth login`) (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods,… - [orl transfers target-org list](/docs/cli/commands/transfers-target-org-list): List orgs you can transfer resources to (needs `orl auth login`), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl upgrade](/docs/cli/commands/upgrade): Upgrade orl to the latest release, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl usage current get](/docs/cli/commands/usage-current-get): Get current-month usage, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl usage daily get](/docs/cli/commands/usage-daily-get): Get per-day spend over a trailing window, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl usage inference-series get](/docs/cli/commands/usage-inference-series-get): Get the org's on-demand inference series, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl vms burn get](/docs/cli/commands/vms-burn-get): Get a VM's burn rate, session cost and runway, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl vms create](/docs/cli/commands/vms-create): Create a VM, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms detail get](/docs/cli/commands/vms-detail-get): Get a VM with GPUs, location, template and price, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl vms disk resize](/docs/cli/commands/vms-disk-resize): Resize a stopped VM's disk, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms fork](/docs/cli/commands/vms-fork): Fork a new VM from a snapshot, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms get](/docs/cli/commands/vms-get): Get a VM by id, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms list](/docs/cli/commands/vms-list): List an org's VMs (cursor-paginated), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms log get](/docs/cli/commands/vms-log-get): Get a VM's console log tail, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms metrics get](/docs/cli/commands/vms-metrics-get): GPU utilization, memory, power and temperature for a pod (unavailable on a VM's passthrough GPUs), part of the orl CLI for OpenRelay GPU VMs, pods, batches… - [orl vms reboot](/docs/cli/commands/vms-reboot): Reboot a VM (beta, not yet executed), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms restart](/docs/cli/commands/vms-restart): Restart a VM, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms snapshot create](/docs/cli/commands/vms-snapshot-create): Create a snapshot of a VM, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms ssh](/docs/cli/commands/vms-ssh): Open an SSH session to a VM over the data plane, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl vms ssh-key attach](/docs/cli/commands/vms-ssh-key-attach): Attach an org SSH key to a VM, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms ssh-key detach](/docs/cli/commands/vms-ssh-key-detach): Detach an org SSH key from a VM (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl vms ssh-key get](/docs/cli/commands/vms-ssh-key-get): SSH keys attached to a VM + the org's keys, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl vms stop](/docs/cli/commands/vms-stop): Stop a VM, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl vms terminate](/docs/cli/commands/vms-terminate): Terminate a VM (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl vms visibility set](/docs/cli/commands/vms-visibility-set): Set a VM's endpoint visibility, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl webhooks create](/docs/cli/commands/webhooks-create): Create a webhook (secret returned once), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl webhooks delete](/docs/cli/commands/webhooks-delete): Delete a webhook (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl webhooks delivery list](/docs/cli/commands/webhooks-delivery-list): Recent delivery attempts for a webhook, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl webhooks list](/docs/cli/commands/webhooks-list): List an org's webhooks (no secrets), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl webhooks secret regenerate](/docs/cli/commands/webhooks-secret-regenerate): Regenerate a webhook secret (returned once) (asks for confirmation; --yes to skip), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. - [orl webhooks toggle](/docs/cli/commands/webhooks-toggle): Enable/disable a webhook, part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example. - [orl webhooks update](/docs/cli/commands/webhooks-update): Update a webhook (name/url/events), part of the orl CLI for OpenRelay GPU VMs, pods, batches and billing. Usage, every flag and an example.