codai Agents API

openapi: 3.2.0
info:
  title: Codai Ro Agents API
  version: '1'
  description: 'Operations tagged Agents across 2 of this provider''s published API definitions: codai-ro-gateway-openapi.json, codai-ro-gateway-openapi.yml. Each path carries the servers of the definition it was published in.'
servers:
- url: https://ai.codai.ro
  description: Production
security:
- bearerAuth: []
tags:
- name: Agents
  description: 'Task-oriented agent API: submit a goal and the gateway runs the whole agent loop server-side (routing, gateway

    tools, sub-agents, memory). `POST /v1/agents/run` blocks until the answer; `POST /v1/agents/runs` returns a run

    id immediately and persists the run so it can be polled, streamed and cancelled. Requires a tier with sub-agents

    and tools (Pro or above).'
paths:
  /v1/agents/run:
    post:
      operationId: runAgentSync
      tags:
      - Agents
      summary: Run an agent task synchronously
      description: 'Submits a goal and blocks until the agent loop finishes, returning the final answer in one response. The

        `task` becomes the user turn (with `context` appended under a `## Context` heading), `system` optionally

        prepends a system message, and the agent-mode scaffold (tool guidance, stop conditions) is injected on top.

        Requires a key whose tier supports sub-agents and tools, otherwise 403. Bills exactly like an agent-mode chat

        turn: every call is a fresh task, so the task-cap chain applies and an exhausted quota is a 429

        `quota_exceeded`. `max_steps` is validated but currently ignored — the loop depth is the tier''s agent-mode

        depth. Nothing is persisted; if the connection drops the result is lost.'
      parameters:
      - $ref: '#/components/parameters/RequestId'
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/AgentRunRequest'
      responses:
        '200':
          description: The completed run.
          headers:
            x-codai-upstream-model:
              description: Concrete upstream model that produced the final turn.
              schema:
                type: string
            x-codai-event-id:
              description: Usage event id for feedback correlation; absent when no usage row was written.
              schema:
                type: string
                format: uuid
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/AgentSyncResult'
        '400':
          $ref: '#/components/responses/BadRequest'
        '401':
          $ref: '#/components/responses/Unauthorized'
        '402':
          $ref: '#/components/responses/PaymentRequired'
        '403':
          $ref: '#/components/responses/Forbidden'
        '429':
          $ref: '#/components/responses/RateLimited'
        '502':
          $ref: '#/components/responses/UpstreamError'
    servers:
    - url: https://ai.codai.ro
      description: Production
  /v1/agents/runs:
    post:
      operationId: createAgentRun
      tags:
      - Agents
      summary: Create an asynchronous agent run
      description: 'Persists a run in the `queued` state and returns its id immediately (202); execution starts on the same

        instance right away and a sweeper resumes it if the instance dies mid-run. Requires a tier with sub-agents

        (403 otherwise) and a real codai account key — legacy compatibility keys cannot own a persisted run (403).

        `client_request_id` makes the call idempotent per user: a repeat with the same value returns the existing

        run id, again with 202. `max_steps` (stored, default 12) and `budget_seconds` (stored, default 300) are

        accepted but NOT enforced yet. Async runs bypass the task-cap chain: no 429 is raised at submit time and the

        run executes with a task count of 0.'
      parameters:
      - $ref: '#/components/parameters/RequestId'
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/AgentRunCreateRequest'
      responses:
        '202':
          description: Run accepted and queued.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/AgentRunAccepted'
        '400':
          $ref: '#/components/responses/BadRequest'
        '401':
          $ref: '#/components/responses/Unauthorized'
        '402':
          $ref: '#/components/responses/PaymentRequired'
        '403':
          $ref: '#/components/responses/Forbidden'
    servers:
    - url: https://ai.codai.ro
      description: Production
  /v1/agents/runs/stats:
    get:
      operationId: getAgentRunStats
      tags:
      - Agents
      summary: Aggregate statistics over your agent runs
      description: 'Counts and averages over the runs created by the calling user in the last `days` days (default 7, clamped

        to 1–90). `avg_latency_ms` is computed from `started_at` to `finished_at` over finished runs only. Legacy

        compatibility keys own no persisted runs and receive only `window_days` and `total: 0`. A non-numeric `days`

        is not validated and results in a server error.'
      parameters:
      - $ref: '#/components/parameters/RequestId'
      - name: days
        in: query
        required: false
        description: Window length in days, clamped to 1–90.
        schema:
          type: integer
          minimum: 1
          maximum: 90
          default: 7
      responses:
        '200':
          description: Aggregates for the window.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/AgentRunStats'
        '401':
          $ref: '#/components/responses/Unauthorized'
        '402':
          $ref: '#/components/responses/PaymentRequired'
    servers:
    - url: https://ai.codai.ro
      description: Production
  /v1/agents/runs/{id}:
    get:
      operationId: getAgentRun
      tags:
      - Agents
      summary: Poll an agent run
      description: 'Returns the current state of one of your runs: status, result or error, step count, token usage and

        timestamps. A run owned by another account is a 404. `result` is set only when `status` is `completed`,

        `error` only when it is `failed`. Poll this or subscribe to `/stream` — both read the same row.'
      parameters:
      - $ref: '#/components/parameters/RequestId'
      - name: id
        in: path
        required: true
        description: Agent run UUID returned by `POST /v1/agents/runs`.
        schema:
          type: string
          format: uuid
      responses:
        '200':
          description: The run.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/AgentRun'
        '401':
          $ref: '#/components/responses/Unauthorized'
        '402':
          $ref: '#/components/responses/PaymentRequired'
        '404':
          $ref: '#/components/responses/NotFound'
    servers:
    - url: https://ai.codai.ro
      description: Production
  /v1/agents/runs/{id}/steps:
    get:
      operationId: listAgentRunSteps
      tags:
      - Agents
      summary: List the persisted steps of a run
      description: 'Returns the step trace of one of your runs, newest first (`seq` descending). Step rows are returned as

        stored, with camelCase keys (`runId`, `latencyMs`, `createdAt`). Today the executor persists exactly ONE

        step per run — a `seq: 0` `answer` step written when the run completes, with the answer clipped to 2000

        chars — so the array is empty while the run is in flight and has one element after success; failed and

        cancelled runs have none. A run owned by another account is a 404.'
      parameters:
      - $ref: '#/components/parameters/RequestId'
      - name: id
        in: path
        required: true
        description: Agent run UUID returned by `POST /v1/agents/runs`.
        schema:
          type: string
          format: uuid
      responses:
        '200':
          description: The step trace.
          content:
            application/json:
              schema:
                type: object
                required:
                - id
                - steps
                properties:
                  id:
                    type: string
                    format: uuid
                    description: The run id.
                  steps:
                    type: array
                    items:
                      $ref: '#/components/schemas/AgentRunStep'
        '401':
          $ref: '#/components/responses/Unauthorized'
        '402':
          $ref: '#/components/responses/PaymentRequired'
        '404':
          $ref: '#/components/responses/NotFound'
    servers:
    - url: https://ai.codai.ro
      description: Production
  /v1/agents/runs/{id}/stream:
    get:
      operationId: streamAgentRun
      tags:
      - Agents
      summary: Stream a run's progress over SSE
      description: 'Server-Sent Events view of one of your runs. The gateway polls the run every second and emits:


        - `event: step` — `{ "seq": number, "kind": "plan"|"tool_call"|"observation"|"answer", "summary": string|null }`

        for every step with a `seq` greater than the last one sent (in practice one `answer` step on completion).

        - `event: done` — `{ "status": "completed"|"failed"|"cancelled", "result": string|null, "error": string|null,

        "step_count": number }` when the run reaches a terminal state; the stream then closes.

        - `event: timeout` — `{}` after 30 minutes on a single connection; the stream closes and you reconnect.


        There are NO heartbeats or comment lines between polls, NO `id:` lines, and `Last-Event-ID` is never read:

        a reconnect is a full replay from `seq` 0. The response is `text/event-stream` with `Cache-Control: no-cache`

        and `x-accel-buffering: no`. A run owned by another account is a 404 before the stream starts.'
      parameters:
      - $ref: '#/components/parameters/RequestId'
      - name: id
        in: path
        required: true
        description: Agent run UUID returned by `POST /v1/agents/runs`.
        schema:
          type: string
          format: uuid
      responses:
        '200':
          description: SSE stream; see the description for event names and payloads.
          content:
            text/event-stream:
              schema:
                type: string
        '401':
          $ref: '#/components/responses/Unauthorized'
        '402':
          $ref: '#/components/responses/PaymentRequired'
        '404':
          $ref: '#/components/responses/NotFound'
    servers:
    - url: https://ai.codai.ro
      description: Production
  /v1/agents/runs/{id}/cancel:
    post:
      operationId: cancelAgentRun
      tags:
      - Agents
      summary: Cancel a queued or running run
      description: 'Marks one of your runs `cancelled`, sets `finished_at` and releases the resume lease. Only `queued` and

        `running` runs can be cancelled; a terminal run, an unknown id or another account''s run all answer 404

        ("not found or not cancellable"). Cancellation is a status flip: an execution already in flight keeps running

        on the instance until the model loop ends, but its result is discarded because the completion update only

        applies to a run still in `running`. Tokens consumed until then are still billed.'
      parameters:
      - $ref: '#/components/parameters/RequestId'
      - name: id
        in: path
        required: true
        description: Agent run UUID returned by `POST /v1/agents/runs`.
        schema:
          type: string
          format: uuid
      responses:
        '200':
          description: The run is now cancelled.
          content:
            application/json:
              schema:
                type: object
                required:
                - id
                - status
                properties:
                  id:
                    type: string
                    format: uuid
                  status:
                    type: string
                    enum:
                    - cancelled
        '401':
          $ref: '#/components/responses/Unauthorized'
        '402':
          $ref: '#/components/responses/PaymentRequired'
        '404':
          $ref: '#/components/responses/NotFound'
    servers:
    - url: https://ai.codai.ro
      description: Production
components:
  responses:
    Unauthorized:
      description: Missing, malformed, revoked or paused API key.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
    Forbidden:
      description: The key is valid but not allowed to perform this action (model allow-list, role, ownership).
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
    PaymentRequired:
      description: The subscription behind this key is inactive.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
    NotFound:
      description: No such resource, or it belongs to another account.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
    RateLimited:
      description: Per-key RPM/RPH, token-per-minute budget, spend cap (`rate_limit_exceeded`) or task quota (`quota_exceeded`). Read `Retry-After`.
      headers:
        Retry-After:
          description: Seconds to wait before retrying.
          schema:
            type: integer
            minimum: 1
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
    BadRequest:
      description: Malformed body or query; `details` holds the flattened field errors.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
    UpstreamError:
      description: Every upstream lane failed (or returned an empty completion) after failover.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
  schemas:
    AgentUsage:
      type: object
      title: Token usage
      description: Aggregate token counts for the whole run.
      required:
      - prompt_tokens
      - completion_tokens
      properties:
        prompt_tokens:
          type: integer
          minimum: 0
        completion_tokens:
          type: integer
          minimum: 0
    AgentRunAccepted:
      type: object
      title: Run accepted
      description: Response of `POST /v1/agents/runs` (HTTP 202).
      required:
      - id
      - status
      - poll
      properties:
        id:
          type: string
          format: uuid
          examples:
          - 3f9d2c1e-7b4a-4e0f-9a1d-2c5b6e7f8a90
        status:
          type: string
          enum:
          - queued
          description: Always `queued` at submit time, even when an idempotent repeat hits a run that has since progressed.
        poll:
          type: string
          description: Relative URL of the poll endpoint for this run.
          examples:
          - /v1/agents/runs/3f9d2c1e-7b4a-4e0f-9a1d-2c5b6e7f8a90
    AgentRunStats:
      type: object
      title: Agent run statistics
      description: 'Aggregates over your runs in the window. Only `window_days` and `total` are guaranteed; the remaining

        fields are absent for legacy compatibility keys that cannot own persisted runs.

        '
      required:
      - window_days
      - total
      properties:
        window_days:
          type: integer
          minimum: 1
          maximum: 90
        total:
          type: integer
          minimum: 0
        completed:
          type: integer
          minimum: 0
        failed:
          type: integer
          minimum: 0
        cancelled:
          type: integer
          minimum: 0
        in_flight:
          type: integer
          minimum: 0
          description: Runs currently `queued` or `running`.
        prompt_tokens:
          type: integer
          minimum: 0
        completion_tokens:
          type: integer
          minimum: 0
        avg_steps:
          type: integer
          minimum: 0
          description: Rounded average of `step_count`.
        avg_latency_ms:
          type: integer
          minimum: 0
          description: Rounded average of `finished_at - started_at` over finished runs.
    AgentSyncResult:
      type: object
      title: Synchronous run result
      description: Response of `POST /v1/agents/run`.
      required:
      - result
      - status
      - usage
      - model
      - event_id
      properties:
        result:
          type: string
          description: The final assistant answer (empty string when the model produced no content).
        status:
          type: string
          enum:
          - completed
        usage:
          $ref: '#/components/schemas/AgentUsage'
        model:
          type: string
          description: Concrete upstream model that served the final turn.
        event_id:
          type:
          - string
          - 'null'
          format: uuid
          description: Usage event id for feedback correlation; `null` when none was written.
    Error:
      type: object
      required:
      - error
      properties:
        error:
          type: object
          required:
          - message
          - type
          - code
          properties:
            message:
              type: string
              description: Human-readable, provider names redacted.
            type:
              type: string
              enum:
              - invalid_request_error
              - subscription_required
              - rate_limit_error
              - upstream_error
              - server_error
            code:
              type: string
              enum:
              - bad_request
              - invalid_api_key
              - subscription_inactive
              - forbidden
              - not_found
              - payload_too_large
              - rate_limit_exceeded
              - quota_exceeded
              - upstream_error
              - internal_error
              - not_a_member
              - lease_held
              - seq_conflict
              - host_offline
              - host_timeout
              - rules_refused
              - search_unavailable
              - task_not_confirmable
              - slug_taken
              - project_has_environments
              - already_has_env
              - not_enrollable
              - not_byo
              - invalid_transition
              - already_member
              - vm_per_dev_single_member
              - last_owner
              - insufficient_balance
              - key_scope_forbidden
              - assets_unavailable
              - asset_not_found
              - too_many_renders
              - renders_unavailable
              - timeline_too_long
              - not_a_timeline
              - invalid_cursor
              - templates_unavailable
              - url_blocked
              - invalid_tokens
              - unsupported_media
              - kind_mismatch
              - node_not_found
              - invalid_pick
              - duplicate_key
              - unknown_var
              - invalid_var_value
              - unsupported_kind
              - too_long
              - stale
              - not_supported_for_kind
              - unsupported_format
              - too_large
              - not_implemented
              - version_conflict
              - settings_unavailable
              - ai_label_required
              - sandbox_concurrency_limit
              - sandbox_busy
              - sandbox_daily_limit
              - invalid_ttl
              - workspace_too_large
              - sandbox_paused_limit
              - sandbox_capacity
            details:
              description: Present on `bad_request` (flattened Zod field errors), `invalid_tokens` (`issues`) and `rules_refused` (`tool`, `rule_id`, `decision`). Other codes never forward details over HTTP.
              type: object
              additionalProperties: true
      examples:
      - error:
          message: Invalid or missing API key.
          type: invalid_request_error
          code: invalid_api_key
    AgentRunStep:
      type: object
      title: Agent run step
      description: One persisted step, as stored (camelCase keys).
      required:
      - id
      - runId
      - seq
      - kind
      - summary
      - detail
      - latencyMs
      - createdAt
      properties:
        id:
          type: string
          format: uuid
        runId:
          type: string
          format: uuid
        seq:
          type: integer
          minimum: 0
          description: Monotonic step index within the run.
        kind:
          type: string
          enum:
          - plan
          - tool_call
          - observation
          - answer
        summary:
          type:
          - string
          - 'null'
          description: Assistant content, tool name or observation summary (redacted, ≤ 2000 chars).
        detail:
          description: Structured detail (tool args/result, redacted). Always `null` for steps written by the current executor.
          nullable: true
        latencyMs:
          type:
          - integer
          - 'null'
        createdAt:
          type: string
          format: date-time
    AgentRunCreateRequest:
      allOf:
      - $ref: '#/components/schemas/AgentRunRequest'
      - type: object
        title: Async agent run request
        description: Body of `POST /v1/agents/runs`; extends the synchronous request.
        properties:
          budget_seconds:
            type: integer
            minimum: 10
            maximum: 1800
            description: Requested wall-clock budget. Stored (default 300) but not enforced.
          client_request_id:
            type: string
            maxLength: 200
            description: Idempotency key scoped to your user — a repeat returns the existing run id.
            examples:
            - ci-build-8812-step-3
    AgentRunStatus:
      type: string
      description: Run lifecycle — `queued` → `running` → one of `completed`, `failed`, `cancelled`.
      enum:
      - queued
      - running
      - completed
      - failed
      - cancelled
    AgentRunRequest:
      type: object
      title: Agent run request
      description: Body of `POST /v1/agents/run`.
      required:
      - task
      properties:
        task:
          type: string
          minLength: 1
          maxLength: 64000
          description: The goal, in natural language. Becomes the user turn.
          examples:
          - Find the three most recent BNR EUR/RON rates and summarise the trend.
        context:
          type: string
          maxLength: 256000
          description: Optional supporting material, appended to the task under a `## Context` heading.
        system:
          type: string
          maxLength: 32000
          description: Optional system-prompt override, sent as the first message before the agent scaffold.
        model:
          type: string
          maxLength: 128
          default: codai
          description: Model alias to route through.
          examples:
          - codai
        max_steps:
          type: integer
          minimum: 1
          maximum: 16
          description: Requested agent-loop depth cap. Validated but not enforced — the tier's agent-mode depth applies.
    AgentRun:
      type: object
      title: Agent run
      description: Poll view of a persisted run (`GET /v1/agents/runs/{id}`).
      required:
      - id
      - status
      - task
      - model
      - result
      - error
      - step_count
      - usage
      - created_at
      - started_at
      - finished_at
      properties:
        id:
          type: string
          format: uuid
        status:
          $ref: '#/components/schemas/AgentRunStatus'
        task:
          type: string
          description: The submitted goal, verbatim.
        model:
          type:
          - string
          - 'null'
          description: Requested model alias (`codai` when omitted at submit).
        result:
          type:
          - string
          - 'null'
          description: Final answer; set only once `status` is `completed`.
        error:
          type:
          - string
          - 'null'
          description: Failure message (≤ 2000 chars); set only when `status` is `failed`.
        step_count:
          type: integer
          minimum: 0
          description: Loop steps executed so far. Not incremented by the current executor.
        usage:
          $ref: '#/components/schemas/AgentUsage'
        created_at:
          type: string
          format: date-time
        started_at:
          type:
          - string
          - 'null'
          format: date-time
          description: When execution was claimed; `null` while `queued`.
        finished_at:
          type:
          - string
          - 'null'
          format: date-time
          description: Set on `completed`, `failed` and `cancelled`.
  parameters:
    RequestId:
      name: x-request-id
      in: header
      required: false
      description: Client correlation id; echoed back as `x-codai-trace-id` and persisted on the usage row.
      schema:
        type: string
        maxLength: 128
  securitySchemes:
    bearerAuth:
      type: http
      scheme: bearer
      bearerFormat: codai_…
      description: A codai API key (`codai_` prefix). Ephemeral tokens from `POST /v1/tokens` are accepted only by the media surfaces matching their scope (`/v1/realtime`, `/v1/audio/*`, `/v1/embeddings`).
x-refined-from:
- codai-ro-gateway-openapi.json
- codai-ro-gateway-openapi.yml