> ## Documentation Index > Fetch the complete documentation index at: https://docs.reactor.inc/llms.txt > Use this file to discover all available pages before exploring further. > ## Agent Instructions > To build and serve your own model, start at /deploy/development/quickstart and /deploy/development/overview. Deploying is the default path: reactor init scaffolds a workspace, reactor auth login authenticates, and reactor model deploy registers the model, publishes the release with the weights/ folder, and activates it on Reactor's GPUs, in one command from that workspace. Docker must be running, because the publish step builds the image locally. Bump model.version in reactor.yaml before redeploying a change, because a release that already has an image is reactivated as it is. Deployment access is granted per account, so contact team@reactor.inc if a deploy is refused. Every key in reactor.yaml is documented at /deploy/platform/reactor-yaml. Model code imports reactor_runtime; Python client code imports reactor_sdk. The runtime overview explains the model interface. Running the model on your own machine with reactor run is optional and needs a GPU you attach with --gpus; /deploy/development/local-testing covers that loop and pairs a complete brightness model with a Python client test in a separate brightness-test workspace. > Reactor hosts multiple models, each with its own connect slug (modelName) and command/event schema. The video model catalog — slug, typed SDK package, and links to its schema — is at /model-api-reference/overview. Robotics policy documentation starts at /robotics/overview; The launch policies are Cosmos Nano DROID (/robotics/cosmos/nano-policy-droid/), DreamZero DROID (/robotics/dreamzero/droid/), and FLUX Action DROID (/robotics/flux-action/droid/), organized by model family and embodiment. Some models expose one slug per experience (e.g. HappyOyster); always take the slug from the model's own pages, never guess it. > Fastest path to a working app: `npx create-reactor-app my-app --model=` scaffolds a complete app with secure auth wired up. Typed TypeScript SDKs are published as @reactor-models/; Python uses the base reactor-sdk package. > Auth: exchange an API key (rk_...) for a JWT via POST https://api.reactor.inc/tokens from your server. Never put the API key in client-side code. > Append .md to any docs URL for clean Markdown. Search these docs via the MCP server at https://docs.reactor.inc/mcp. # Deploy a release > Deploys a registered release of a model you own, for example to roll back or to activate a release registered with `deploy: false`. The CLI equivalent is `reactor model deploy`. `instances` is the full desired instance plan, one entry per region. A region you omit from a non-empty plan is scaled to zero, and `instances: []` scales every region to zero. Omitting `instances` leaves capacity unchanged. For a GPU model, the plan is checked against your account's accelerator allocation in each region. Valid region names come from `GET /clusters/regions`; the plan may hold at most 1000 instances. `strategy` is `ROLLING` (default; instances are replaced one at a time) or `REPLACE_ALL` (the whole old generation stops at once; faster, but a release that never becomes ready leaves the model with no capacity). Values are matched exactly. Neither strategy ends a running session early. `immediate: true`, legal only with `REPLACE_ALL`, also closes the model's live sessions once the deploy is sent. A release with no weights of its own reuses the weights of the most recent deployed release that has a complete bundle (`weights_carried_from`), or deploys without weights (`weights_missing: true`). Callable by an `account_admin` or `developer` of the owning account when model serving is enabled for the account. ## OpenAPI ````yaml /openapi.yaml put /models/{id}/releases openapi: 3.1.0 info: title: Reactor Platform API version: '1' description: > The Reactor Platform API mints session tokens, starts and operates sessions, negotiates media transport, and manages your account, models, and deployments. Every path is relative to `https://api.reactor.inc`. See the [Introduction](/platform-api/introduction) and [Authentication](/platform-api/authentication) pages for credentials, versioning, and errors. contact: name: Reactor support url: https://docs.reactor.inc/resources/support servers: - url: https://api.reactor.inc description: Production security: - apiKey: [] tags: - name: Status description: >- Public, unauthenticated endpoints for service health, pricing, and the values the API accepts. - name: Tokens description: > Exchange a long-lived API key for a short-lived, session-scoped token that a client can use to create and operate sessions without holding the key. - name: Users description: > Read your own user record, your account's roster, and account memberships, and manage member roles as an account admin. - name: API Keys description: >- Revoke API keys. Use the [Dashboard](https://reactor.inc/dashboard) to list or create keys. The CLI can also create a key with `reactor auth login`. - name: Sessions description: > Create, inspect, and end sessions. A session holds one running model for one client application; media and messages flow over a transport you attach to it afterwards. - name: Transport description: > Attach a client to a session. WebRTC signaling carries ICE servers, SDP, and ICE candidates; a LiveKit room connection lets the session join a room you host. - name: Files description: > Upload files a model reads, either for one running session or ahead of time for sessions you have not started yet. - name: Recordings description: > Fetch the recorded video of a session as an HLS playlist, during the session or after it ends. - name: Accounts description: > Read and manage the account your credential acts in: its name and slug, its credit balance, its effective quotas, and its accelerator capacity. - name: Invitations description: > Invite people to your account by email, and manage pending invitations. The recipient accepts from the emailed link while signed in to the dashboard. - name: Usage description: > List the sessions your account has run, filter them by status, date, model, or API key, and read exact session totals for a date window. - name: Models description: > Register, read, update, and delete the models your account serves, and list the models you can use. The `reactor model` CLI commands are built on these endpoints. - name: Releases description: > Register a model's versioned releases, push their container image, and deploy a release with a per-region instance plan. - name: Model Weights description: > Upload and inspect the weights bundle of a release: single-request uploads, multipart uploads for large files, and the server-side finalize that marks a bundle complete. - name: Model Schemas description: > Attach an OpenAPI schema to a release that describes the model's commands, messages, and tracks. Schemas of public models are readable without credentials. - name: Model Access description: | Share a private model with other accounts, one environment at a time. - name: Projects description: > A project is the deploy identity of a model you serve on Reactor. Each model is one project, with its own UUID. Creating a project also creates its `production` environment. - name: Environments description: > An environment is a named deploy target of a project, such as `production` or `staging`. Each write to an environment creates its next revision, and the newest revision is what the environment runs. Environment access grants let another account start sessions on one environment. - name: Revisions description: > A revision is an immutable, numbered snapshot of one environment: its model image, weights, resources, instances, and variables. An environment's revisions, newest first, are its change log. - name: Images description: > Model images are the container images pushed to your model's repository, named by digest. Experimental: this API can change without notice. - name: Weight Sets description: > A weight set is a version-2 weights bundle, named by its set digest. Push the files, then put the manifest. Available when weight sets are enabled for your account. Experimental: this API can change without notice. - name: Tags description: > A tag is a movable name on one model image or one weight set of a project. A release of the model is a read-only tag. Experimental: this API can change without notice. paths: /models/{id}/releases: parameters: - $ref: '#/components/parameters/ModelId' put: tags: - Releases summary: Deploy a release description: > Deploys a registered release of a model you own, for example to roll back or to activate a release registered with `deploy: false`. The CLI equivalent is `reactor model deploy`. `instances` is the full desired instance plan, one entry per region. A region you omit from a non-empty plan is scaled to zero, and `instances: []` scales every region to zero. Omitting `instances` leaves capacity unchanged. For a GPU model, the plan is checked against your account's accelerator allocation in each region. Valid region names come from `GET /clusters/regions`; the plan may hold at most 1000 instances. `strategy` is `ROLLING` (default; instances are replaced one at a time) or `REPLACE_ALL` (the whole old generation stops at once; faster, but a release that never becomes ready leaves the model with no capacity). Values are matched exactly. Neither strategy ends a running session early. `immediate: true`, legal only with `REPLACE_ALL`, also closes the model's live sessions once the deploy is sent. A release with no weights of its own reuses the weights of the most recent deployed release that has a complete bundle (`weights_carried_from`), or deploys without weights (`weights_missing: true`). Callable by an `account_admin` or `developer` of the owning account when model serving is enabled for the account. operationId: deployRelease requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/ModelDeployRequest' examples: plan: summary: Deploy with a two-region plan value: release: v1.2.3-gabcdef0 instances: - region: us-west count: 2 - region: us-east count: 1 rollback: summary: Roll back without touching capacity value: release: v1.2.2-g1234567 responses: '200': description: The deploy was accepted. content: application/json: schema: $ref: '#/components/schemas/ModelDeployResponse' example: release: v1.2.3-gabcdef0 weights_carried_from: v1.2.2-g1234567 immediate: false revision: number: 3 '400': description: > `release` is missing or not semver-style, `strategy` is unknown, `immediate` is set without a `release` or without `REPLACE_ALL`, the plan is invalid (`invalid_deployment_plan`: unknown or duplicate region, negative count, more than 1000 instances), or an environment variable is invalid. content: application/json: schema: $ref: '#/components/schemas/Error' '401': $ref: '#/components/responses/Unauthorized' '403': description: >- You do not own the model (returned whether or not it exists), or model serving is not enabled for your account. content: application/json: schema: $ref: '#/components/schemas/Error' '404': description: The release tag is not registered for this model. content: application/json: schema: $ref: '#/components/schemas/Error' '409': $ref: '#/components/responses/ModelDeployConflict' '429': description: The deploy exceeds one of your account's resource quotas. content: application/json: schema: $ref: '#/components/schemas/Error' '503': description: > A dependency is temporarily unavailable. `error` is `accelerator_inventory_unavailable` when a planned region's capacity could not be read. Retry the same request. content: application/json: schema: $ref: '#/components/schemas/Error' components: parameters: ModelId: name: id in: path required: true description: >- The model's ID (not its project ID). Find it with `GET /models?name=your-org/your-model`. schema: type: string format: uuid schemas: ModelDeployRequest: type: object properties: release: type: string description: The release to deploy. Required on `PUT`; optional on `PATCH`. example: v1.2.3-gabcdef0 instances: type: array description: >- The instance plan per region. On `PUT` it replaces the plan; on `PATCH` it merges. Omit to leave capacity unchanged. items: type: object required: - region - count properties: region: type: string description: A region from `GET /clusters/regions`. example: us-west count: type: integer minimum: 0 strategy: type: string enum: - ROLLING - REPLACE_ALL default: ROLLING description: How the old release stops serving. Matched exactly. immediate: type: boolean default: false description: >- Close the model's live sessions once the deploy is sent. Requires a `release` and `REPLACE_ALL`. ModelDeployResponse: type: object required: - release - immediate properties: release: type: string description: The release now deployed. weights_carried_from: type: string description: The release whose weights were reused because this release has none. weights_missing: type: boolean description: >- `true` when the release deployed with no weights available. Omitted when `false`. immediate: type: boolean description: Whether this deploy closes live sessions. revision: type: object description: >- The `production` revision this deploy created. Omitted when nothing changed. properties: number: type: integer Error: type: object required: - error properties: error: type: string description: >- A stable error code (for example `not_found`) on newer routes, or a sentence on older ones. When `message` is present, branch on `error`. message: type: string description: >- A human-readable explanation. Do not parse it; its wording may change. resource: type: string description: The type of the resource a `not_found` refers to. reason: type: string description: >- A more specific cause for `failed_precondition`, `aborted`, and some `invalid_request` errors. responses: Unauthorized: description: The credential is missing, invalid, or expired. content: application/json: schema: $ref: '#/components/schemas/Error' ModelDeployConflict: description: > The deploy cannot be applied; nothing changed. `error` is one of: - `accelerator_not_supported_in_region`: a planned region cannot run the model's accelerator type. The message names the regions that can. - `accelerator_quota_exceeded`: the plan exceeds your account's allocation of the accelerator type in a region. - `failed_precondition` with `reason` `deployment_plan_unplaceable`: a region has too few free accelerators right now. Retry later or lower the count. - `failed_precondition` with `reason` `accelerators_required`: the GPU model has no `gpu_type`. Set it with `PUT /models/{id}`. - `weights_bundle_unverifiable`: the release has no weights of its own and the bundle it would reuse is too large to verify. Upload this release's weights. - `aborted` with `reason` `concurrent_update`: another write changed the deployment. Retry. A sentence in `error` with no `message` means the release has no recorded image, or (on `PATCH` without a release) the model has no deployed release to scale. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: accelerator_quota_exceeded message: >- deployment plan would commit 12 'NVIDIA_H100' accelerators in region 'us-east', but only 10 are allocated to this account. Lower the requested count or request a larger allocation securitySchemes: apiKey: type: http scheme: bearer description: >- A Reactor API key (`rk_...`), sent as `Authorization: Bearer rk_...`. Keep it on a server. ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.