> ## Documentation Index > Fetch the complete documentation index at: https://docs.judt.ai/llms.txt > Use this file to discover all available pages before exploring further. # Send a message > Starts a turn with your message. If the session already has an active turn, the new turn waits in a queue behind it. The queue holds up to five turns, and they run in the order that you sent them. The response returns when the turn starts or joins the queue, not when it finishes. To follow the turn, stream the session's events or get the turn. ## OpenAPI ````yaml /api-reference/openapi.json post /api/apps/{id}/sessions/{session_id}/turns openapi: 3.1.0 info: title: Judt API version: 1.0.0 description: >- The Judt API creates and runs Apps. Each App has a sandbox, where you and agents edit and run its source, and a production deployment that serves it. ## Authentication Send an API token or an OAuth access token in the `Authorization` header. - **API token:** send it as a bearer token. A token belongs to one organization, and you create and revoke tokens in the dashboard. A token scoped to one App can reach only that App. An organization-wide token can also create a short-lived token for one App with `POST /api/tokens`, to give to an agent or an MCP client. - **OAuth access token:** an agent gets one when a person approves its access to one organization, so the person never handles a token. Reads need the `apps:read` scope, and changes need `apps:write`. To register and sign in, follow https://judt.ai/auth.md. Send the token as a bearer token, or with a DPoP proof if it's bound to a DPoP key. One access token can work on both this API and the MCP server at https://api.judt.ai/mcp. ## Sandboxes Creating an App reserves its address, but it doesn't start a sandbox or publish anything. There's no endpoint to start a sandbox. A request that needs the sandbox starts it first, so the first request after the sandbox goes idle takes longer. Files, Git, commands, commits, restores, previews, and deployments all need the sandbox. The sandbox stops on its own when it's idle. If `sandbox_status` is `failed`, the most recent start failed, and the API retries it while the sandbox is needed. ## Pagination A list returns one page at a time, in the form `{"results": [...], "has_more": true, "next_cursor": "CURSOR"}`. To get the next page, list again with `cursor` set to `next_cursor`, and the same `sort` and filters. Results that arrive while you page don't shift your pages. On the last page, `has_more` is `false`, and `next_cursor` still finds results added later. ## Retries A request that creates, changes, or deletes something returns `200` after the work is done, with the resource as it is afterward. To retry a request safely, send an `Idempotency-Key` header. For 30 days, a retry with the same key and input returns the first request's result instead of repeating the work. If you don't send a key, the API generates one and returns it in the `Idempotency-Key` response header. Setting and deleting a Secret take no key, because repeating them changes nothing. A visitor token or a preview link is created fresh each time, because nothing is stored to repeat. The API records file writes and deletes, commands, commits, and restores before it runs them. The response status tells you what to do next: - `503` or `504`: the action is recorded but not finished. To get its result, retry with the same key. A new key would run the action again. - `4xx` or `502`: the result is final. A retry with the same key returns the same response. A `504 request_timeout` response says how far the work got and which event to check. The work continues, and a retry with the same key picks it up. ## Troubleshooting To see what happened to an App's sandbox, deployments, and deletion, list its events with `GET /api/apps/{id}/events`. To see what the App's code did, including requests, output, exceptions, and failed builds, list its logs with `GET /api/apps/{id}/logs`. ## Time limits Requests have the following time limits: - Stopping the sandbox: 120 seconds. - File writes, commits, and restores: 120 seconds. - Commands: the command's `timeout_seconds`, up to 120 seconds, plus 60 seconds to stop it and save its result. - Publishing, unpublishing, deleting an App, and changing production settings: 540 seconds, including any sandbox start. If a file write, commit, restore, or command has to start the sandbox, it gets up to 120 more seconds. ## Errors An error returns a JSON body in the form `{"error": {"code": "CODE", "message": "MESSAGE", "detail": "DETAIL", "request_id": "REQUEST_ID"}}`. The `code` value is stable, so check it rather than `message`. The `detail` value is optional. When a parameter or a body field isn't valid, the request returns `400 invalid_request`, `message` names the field, and `detail` lists every problem if there's more than one. When the server rejects a credential, it returns a JSON error body and sets the `WWW-Authenticate` header. A `401` response includes the `resource_metadata` URL in `WWW-Authenticate` (RFC 9728): | Status | Error code / header | Cause | Action | |---|---|---|---| | `401` | `unauthenticated` | The request has no credential, or the API token or OAuth access token is invalid, expired, or revoked. | Send a valid API token or OAuth access token. | | `401` | `invalid_token` | `WWW-Authenticate` error when an API token or OAuth access token is expired, revoked, or issued for a different `resource` audience. | Refresh the access token, or request a token for the target server. | | `401` | `invalid_dpop_proof` | The `DPoP` proof header is missing, expired, replayed, or signed for a different HTTP method or URL. | Create a new DPoP proof for the request method and URL. | | `403` | `insufficient_scope` | The OAuth token does not have the required scope (`apps:read` or `apps:write`). | Request authorization with the scope named in `WWW-Authenticate`. | | `403` | `app_scoped_token` | An App-scoped API token called `POST /api/apps` or `POST /api/tokens`. | Use an organization-wide API token. | | `403` | `api_token_required` | An OAuth access token called `POST /api/tokens`. | Use an organization-wide API token. | | `403` | `sign_in_required` | A Personal Agent Session Token was used before the user signed in or after the user signed out. | Tell the user to sign in to the Session. | | `403` | `forbidden` | Your organization role (`member`) does not allow this action. | Ask an organization `owner` or `admin` to do the action. | servers: - url: https://api.judt.ai security: [] tags: - name: Account description: >- See who a credential acts for: the person, the organization, and what the credential can reach. - name: Apps description: Create, update, and delete Apps, and control who can reach them. - name: Tokens description: >- Create short-lived API tokens that reach one App, to give to an agent or an MCP client. - name: Secrets description: >- Store values, such as API keys, that an App reads at runtime. Responses never include Secret values. - name: Files description: Read and write the source files in an App's sandbox. - name: Sandbox description: >- Run commands, use Git, commit and restore source, and preview the App's sandbox. - name: Deployments description: Publish an App's source to production, and take it offline. - name: Monitoring description: See what happened to an App, and what its code did. - name: Templates description: The starting points for new Apps. - name: Agents description: >- Work with the in-product coding agent: sessions, turns, events, and the models it runs on. The agent itself works only through this API. - name: Feedback description: Tell the Judt team about bugs, friction, and anything else. paths: /api/apps/{id}/sessions/{session_id}/turns: post: tags: - Agents summary: Send a message description: >- Starts a turn with your message. If the session already has an active turn, the new turn waits in a queue behind it. The queue holds up to five turns, and they run in the order that you sent them. The response returns when the turn starts or joins the queue, not when it finishes. To follow the turn, stream the session's events or get the turn. operationId: create_turn parameters: - name: id in: path required: true description: A unique ID that starts with `app_`. schema: type: string pattern: ^app_[0-9A-Za-z]{24}$ example: app_Zj7XhMgplmTJcp5Dk4vcYuk0 - name: session_id in: path required: true description: A unique ID that starts with `ses_`. schema: type: string pattern: ^ses_[0-9A-Za-z]{24}$ example: ses_Zj7XhMgplmTJcp5Dk4vcYuk0 - name: Idempotency-Key in: header required: false description: >- A key that identifies this request for retries, up to 200 printable characters. A retry with the same key and input returns the first request's result instead of repeating the work. If you omit the key, the API generates one and returns it in the `Idempotency-Key` response header. schema: type: string minLength: 1 maxLength: 200 requestBody: required: true content: application/json: schema: $schema: https://json-schema.org/draft/2020-12/schema type: object properties: message: type: string minLength: 1 maxLength: 32768 description: What to ask the agent, up to 32 KiB. model: description: 'A model for this turn only. Default: the session''s model.' type: string maxLength: 100 effort: type: string enum: - low - medium - high - xhigh - max description: >- How much the model thinks, from `low` to `max`. Higher effort is slower and uses more tokens, and helps on tasks that need careful reasoning. Default: Auto's choice when `model` is `auto`; otherwise the model's default. required: - message additionalProperties: false responses: '200': description: The new turn, running or queued. content: application/json: schema: $ref: '#/components/schemas/Turn' headers: Idempotency-Key: schema: type: string description: >- The key this request ran under: the one you sent, or one the API generated. To pick up the work after a timeout, retry with this key. Idempotency-Retry-Window: schema: type: string example: '2592000' description: >- How long you can retry with the same key, in seconds: 2592000 (30 days). '400': $ref: '#/components/responses/Error' '401': $ref: '#/components/responses/Error' '403': $ref: '#/components/responses/Error' '404': $ref: '#/components/responses/Error' '409': description: >- - `agent_busy`: another session of this App has an active turn, and `detail` names it. An App runs one agent at a time. - `queue_full`: the session already has five queued turns. - `idempotency_conflict`: the key was used with a different request. - `idempotency_expired`: the key is too old to reuse. content: application/json: schema: $ref: '#/components/schemas/Error' '410': $ref: '#/components/responses/Error' '422': description: '- `model_unavailable`: the model is unknown or no longer offered.' content: application/json: schema: $ref: '#/components/schemas/Error' '429': description: >- - `agent_limit`: the organization has reached its limit of running turns. Turns waiting for input do not count. content: application/json: schema: $ref: '#/components/schemas/Error' security: - ApiToken: [] components: schemas: Turn: $schema: https://json-schema.org/draft/2020-12/schema type: object properties: id: type: string description: The turn's ID, which starts with `trn_`. session_id: type: string description: The ID of the session that the turn belongs to. status: type: string enum: - queued - running - waiting - completed - failed - cancelled description: >- The turn's state: - `queued`: waiting behind the session's active turn. - `running`: the agent is working. - `waiting`: the agent needs input from a person, which `input_request` describes. - `completed`, `failed`, or `cancelled`: the turn has ended, and its status doesn't change again. message: type: string description: The message that started the turn. model: type: string description: >- The model that the turn runs on. If `requested_model` is `auto`, this is the model that Auto chose. requested_model: type: string description: 'The model that the turn asked for: `auto` or a model ID.' effort: type: string enum: - low - medium - high - xhigh - max description: >- How much the model thinks on this turn, from `low` to `max`. Auto chooses it by how much reasoning the task needs, unless the request set `effort`; with a chosen model, it is the model's default unless the request set it. input_request: anyOf: - type: object properties: id: type: string description: The input request's ID. kind: type: string const: secrets description: The kind of input that the agent needs. secrets: type: array items: type: object properties: name: type: string description: The Secret's name. provided: type: boolean description: True if the Secret is set; false otherwise. required: - name - provided additionalProperties: false description: >- The Secrets that the agent needs. To provide one, set it with `PUT /api/apps/{id}/secrets/{name}`. The turn resumes after all of them are set. The agent never receives their values. reason: type: string description: Why the agent needs the input, in the agent's words. expires_at: type: string format: date-time description: >- When the request expires. If no one provides or declines the input by then, the turn fails with `input_expired`. example: '2026-10-03T14:01:30.000Z' required: - id - kind - secrets - reason - expires_at additionalProperties: false description: A request for input from a person. - type: 'null' description: >- The input that the agent is waiting for, or null if the turn isn't waiting. base_commit_id: description: >- The commit that the turn started from. To undo the turn, pass this ID as `commit_id` to `POST /api/apps/{id}/restore`. type: - string - 'null' error: anyOf: - type: object properties: code: type: string description: A stable code for the error. message: type: string description: A readable explanation of the error. required: - code - message additionalProperties: false - type: 'null' description: >- Why the turn failed, or null if it didn't fail. The code is one of the following: - `agent_error`: the agent stopped unexpectedly. - `app_deleted`: the App was deleted. - `input_expired`: no one answered the input request in time. - `interrupted`: the turn was interrupted repeatedly without progress. - `turn_budget_exceeded`: the turn ran out of time. - `user_deleted`: the user the turn acts for deleted their account. usage: type: object properties: input_tokens: type: integer minimum: 0 description: The number of tokens the model read. example: 12840 output_tokens: type: integer minimum: 0 description: The number of tokens the model wrote. example: 1260 cache_read_tokens: type: integer minimum: 0 description: The number of input tokens read from the prompt cache. example: 9600 required: - input_tokens - output_tokens - cache_read_tokens additionalProperties: false description: The tokens that the turn used. created_by: description: >- The ID of the user the turn acts for, or null if that user deleted their account. The turn has that user's current role. type: - string - 'null' created_at: type: string format: date-time description: When the turn was created. example: '2026-10-03T14:01:30.000Z' started_at: anyOf: - type: string format: date-time example: '2026-10-03T14:01:30.000Z' - type: 'null' description: When the turn started running, or null if it hasn't started. ended_at: anyOf: - type: string format: date-time example: '2026-10-03T14:01:30.000Z' - type: 'null' description: When the turn ended, or null if it hasn't ended. required: - id - session_id - status - message - model - requested_model - effort - input_request - base_commit_id - error - usage - created_by - created_at - started_at - ended_at additionalProperties: false Error: type: object required: - error properties: error: type: object required: - code - message - request_id properties: code: type: string description: A stable code for the error, such as `not_found`. message: type: string description: A readable explanation of the error. detail: type: string description: >- More information, if any: the sandbox's error, a provider message, or, for a `504`, how far the work got and the ID of its latest event. request_id: type: string description: The request's ID. Include it when you contact support. responses: Error: description: An error, with its code and the request ID. content: application/json: schema: $ref: '#/components/schemas/Error' securitySchemes: ApiToken: type: http scheme: bearer description: >- An API token for one organization. You create and revoke tokens in the dashboard. A token scoped to one App can reach only that App, so give that kind of token to an agent. An organization-wide token can create a short-lived one with `POST /api/tokens`. ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.