> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gloo.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Create a retrieval golden set

> Create a retrieval golden set — golden cases of a query and its expected keys, searched in namespaces the credential can read. Requires knowledge:manage.

{/* forge-v2-compatibility-notes */}

> [!NOTE]
>
> The reference panel describes the current **v3** operation. The notes below describe its **v2 compatibility** API. Follow the v3 request schema for current calls: v3 can move identifiers from the body or query into the URL. Keep the v2 method, path, and payload together when using the compatibility API.

## v2 compatibility notes

`POST /api/v2/knowledge/goldenSets/create` creates a golden set, as a person does on the Console Knowledge Quality page. A run of the set measures how well retrieval finds each case's golden keys. It needs `knowledge:manage` and answers `201` with the `KnowledgeGoldenSet`.

| Body field | Meaning |
| - | - |
| `name` | Required, not blank. |
| `scope` | Required, not blank: a label for what the set measures, such as a team's runbooks. |
| `namespaces` | Required, 1–50 knowledge namespaces each case is searched in. Each must be one the credential can read; otherwise `403`, recorded, with a message naming the namespace. |
| `cases` | Required, 1–100 cases, each `{ caseId, query, expectedKeys }`. `caseId` must not repeat; `expectedKeys` is 1–50 non-empty keys a good retrieval returns for the query (a case key, or a synced entry's source id). |
| `status` | `active` (default) or `archived`. |

The response is `{ id, name, scope, namespaces, cases, status, createdAt, lastRunRequestedAt }`; `lastRunRequestedAt` is when a run was last requested on demand, or `null`. The set is recorded as created by the credential, which is never returned. Namespaces resolve from the credential, never the request; a Personal key acts as its owner.

An unknown field, a blank name or scope, a count outside its range, a repeated or blank `caseId`, a blank query or an unknown status answers `400 bad_request`.

Send an `Idempotency-Key` to retry safely. A repeat returns the recorded `201` only while the credential still holds `knowledge:manage` and `knowledge:read` and still reads every namespace the set searches now and every namespace the recorded response names; otherwise it answers `403` for a missing `knowledge:read`, or `409 idempotency_result_unavailable`. See [Knowledge curation replay](https://docs.gloo.com/forge/reference/idempotency.md#knowledge-curation-replay).

```bash
curl -X POST "$FORGE_BASE/api/v2/knowledge/goldenSets/create" \
  -H "Authorization: Bearer $FORGE_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: create-runbook-golden-set-1" \
  -d '{ "name": "Runbooks", "scope": "Company runbooks", "namespaces": ["company"], "cases": [{ "caseId": "rotate-keys", "query": "How do I rotate the signing keys?", "expectedKeys": ["rotate-signing-keys"] }] }'
```


## OpenAPI

````yaml /forge/reference/openapi-v3.json post /api/v3/knowledge-golden-sets
openapi: 3.1.0
info:
  title: Forge Platform API v3
  version: 0.23.0
  description: >-
    The Forge Platform API in its resource grammar: plural kebab-case
    collections, ids in the path, standard methods for the standard verbs and
    `POST …/{verb}` for the rest, and `/api/v3/me` for the caller's own records.
    Each operation is served by the same handler as the `/api/v2` operation it
    names: its operation id, credentials, permissions, refusals and idempotency
    are that operation's.
  contact:
    name: Forge Platform Team
servers:
  - url: https://api.forge.gloo.com
    description: Production
security:
  - BearerAuth: []
  - PlatformApiKey: []
  - PersonalKey: []
tags:
  - name: Agents
    description: >-
      Agent definitions: CRUD on the definition layer (create, get, list,
      update, delete) plus the A2A AgentCard endpoint, an agent's configuration,
      readiness and open setup request, the template gallery, the delegated work
      agents dispatched (sub-agent task runs and consultations), and agent
      administration: the kill switch, accountability repairs and setup
      requests.
  - name: Skills
    description: >-
      Skill definitions composed into agent prompts. List, fetch, and write
      skills owned by your External Application, and list and manage the skill
      sets that group them.
  - name: Tools
    description: >-
      Safe tool and tool-set catalog metadata and a tool's insights: the agents
      holding it, its calls, usage, history and sandbox rejections; workspace
      tool-set writes; no invocation.
  - name: Models
    description: Read-only access to the model catalog (provider, context window, status).
  - name: Workforce
    description: >-
      Organization units and positions in the caller’s workforce scope closure,
      and pods, pod membership previews and writes, pod lifecycle, ownership,
      internal profiles and Customer assignments, people (create, update, merge,
      offboard, reactivate), the worker directory, worker detail and seat
      budgets, assignment history, position assignment writes and the worker
      role list at the platform scope.
  - name: Objectives
    description: Read-only objective and key-result progress at the deployment root.
  - name: Planning
    description: >-
      Blueprint roadmaps, workstreams and milestones, and the critical-path and
      blocked-task projections, at the deployment root.
  - name: Projects
    description: Read-only project, customer relationship, and objective-link projections.
  - name: Workflows
    description: >-
      Workflow definitions and dispatch. List/get pinned workflow graphs and
      trigger runs by name.
  - name: Runs
    description: >-
      Workflow runs: list and read runs, their steps with runtime content,
      files, timeline, validator passes, recalled context, citations,
      notifications and governance status, the debug bundle, a Markdown export,
      agent conversations, the diagnostic timeline and AI analyses; cancel or
      restart a run, retry a failed notification, start an AI analysis and set a
      run's kill switch.
  - name: Chat
    description: >-
      Hybrid Workforce Chat — channels, sessions, and messages. Read
      channel/session state and post messages.
  - name: Memory
    description: >-
      Agent memory — channel and long-term layers. List, read, and write memory
      entries.
  - name: Knowledge
    description: >-
      Scoped knowledge: vector search, grounded answers, ingestion, entry reads
      and deletion. Every call is limited to the namespaces the credential may
      read or write, resolved server-side.
  - name: Factories
    description: >-
      Factory production units, their workers and tasks, spend and blueprints;
      operating a factory: messaging its manager agent and pausing, resuming or
      archiving it.
  - name: Sandboxes
    description: >-
      A run's sandboxes: recent runs that used them, the cached fleet summary, a
      run's lineage graph, its active sandboxes and a step's commands with
      masked output; diff two snapshots and replay a command against one in a
      fresh VM. Decided at the run's home and companies (factories:read; replay
      factories:manage).
  - name: Governance
    description: >-
      Approvals, policies and recorded decisions. List pending approvals, fetch
      detail, record decisions, list approval policies, list trust policies and
      their revisions, replay a trust policy against recorded decisions
      (historical, never an authorization), explain a recorded Trust Fabric
      decision, page recorded decision evidence, and read the permission catalog
      and role definitions.
  - name: Synthesis
    description: >-
      The synthesis queue: list and read synthesis requests, their Builder
      conversations and the procedures held for operator review, and promote,
      demote, accept or assign a reviewer to up to 50 requests per call. Decided
      at the platform scope.
  - name: Judges
    description: >-
      Machine judges: the judge catalog with metrics and rubrics, drift alerts,
      each judge's verdicts, pending primitive admissions, and the verdicts and
      live judge activity on a skill, agent, procedure, workflow draft or run.
      Decided at the platform scope.
  - name: Evals
    description: Evaluation sets and their runs.
  - name: External apps
    description: >-
      External Application registry: admin-only CRUD for the OAuth2 client
      credentials issued to integrating systems, an app's own record, and the
      agents an External App owns.
  - name: Costs and budgets
    description: >-
      Token spend and budgets. Bucketed and per-customer cost summaries are
      served from cron-populated rollups, so responses carry `meta.asOf`: an
      empty result with a null `asOf` means 'not accrued yet', not 'no spend'.
      Also agent and chat cost aggregates, the budget hierarchy, budget status
      and budget alerts.
  - name: Customers
    description: >-
      End-customer records, organizations and individuals. A customer row is
      shared across apps by `workosId`. The `/api/v2/external/customers` routes
      are an External Application's own view: per-app visibility comes from a
      `customerScopes` link, and those reads are fenced to the calling app. The
      `/api/v2/customers/*` routes are the organization's whole directory,
      decided at the deployment root for any credential holding `customers:read`
      there.
  - name: Procedures
    description: >-
      Read-only Procedures catalog — the deterministic, versioned units a
      workflow graph can bind. Compact projection only: never source code or
      bundled files.
  - name: Health
    description: Authenticated liveness check for integration smoke tests.
  - name: Capabilities
    description: >-
      Read-only deployment configuration: capability domains, their active
      provider chains, the provider catalog, credential health and a
      capability's recent changes. Decided at the deployment root; never
      credential values or option values.
  - name: Actor
    description: >-
      Request-scoped self-discovery: the calling credential's verified actor,
      its current narrowed permissions and each documented operation's advisory
      prerequisites.
  - name: API keys
    description: >-
      Personal keys and host sign-ins: a key owner lists and revokes their own;
      an administrator lists and revokes any person's.
  - name: Connections
    description: >-
      Connection credentials and the credential registry: list, create, review,
      revoke and delete credentials, and the vendors and connections they bind.
  - name: Email
    description: >-
      Inbound and outbound email: the outbox, received mail, the inbound ingress
      and its routes, and the sending domain.
  - name: Environments
    description: >-
      Environment Connections and their credentials: list, create, rename,
      replace the credential, and delete.
  - name: Identity
    description: >-
      Person profiles and avatars, and unmapped sign-in attempts waiting to be
      mapped or dismissed.
  - name: Notifications
    description: >-
      Notification event types with their payload fields, and the outbound Slack
      sends.
  - name: Observe and ops
    description: >-
      Operational reads: attention items, investigation notes, workflow
      performance and reliability overviews, failed runs, slow steps and a run's
      audit events.
  - name: Org units and positions
    description: >-
      The org chart: drafts that stage org unit and position changes, and the
      org units, positions and assignments they publish.
  - name: People and members
    description: >-
      People, their worker seats and org unit memberships, expertise tags, and
      the person-identity integrity check.
  - name: Projects and staffing
    description: Intake requests, their plans, and tasks across projects.
  - name: Roles and grants
    description: >-
      Job roles, the tools and skills they grant or deny, a holder's role
      grants, and AI-drafted role content.
  - name: Trust Fabric
    description: >-
      Governance controls and evidence: decisions, command tiers, trust
      assurance, policies, the governance door, the publish judge and the
      platform kill switch.
  - name: Vendors
    description: >-
      Vendors and their connections, attached API specs, generated client
      surfaces and remote MCP tool catalogs.
  - name: Webhooks
    description: >-
      Webhook subscriptions, deliveries and their attempts, replay and
      redelivery, dead letters, delivery health and the canonical event log.
  - name: Files
    description: >-
      Raw-byte file upload and authorized content redirects for uploaded files
      and readable run outputs.
  - name: Compliance
    description: >-
      Deployment-wide audit and governance record exports and a decision
      summary, for auditors holding `audit:read` at the deployment root.
paths:
  /api/v3/knowledge-golden-sets:
    post:
      tags:
        - Knowledge
      summary: Create a retrieval golden set
      description: >-
        Requires permission `knowledge:manage`.


        Creates a golden set over namespaces the caller may read (every
        namespace it names must be readable, so that every namespace a later run
        searches is one the caller may see), recorded as created by the
        credential. A run measures how well retrieval finds each case's golden
        keys. Who created a set and the embeddings URL a run measured are never
        returned. With an `Idempotency-Key`, a repeat of a completed request
        returns the recorded response (`Idempotency-Replayed: true`) without
        writing or scheduling again, only while the caller still holds
        `knowledge:manage` and `knowledge:read` and still reads every namespace
        the golden set the recorded response names searches now and, for a
        create or update, every namespace the recorded set names; otherwise it
        answers 403 for a missing `knowledge:read` or the non-disclosing 409
        `idempotency_result_unavailable`. A repeat of a refused request (400,
        403, 404, 409) answers that 409 too, not the refusal again.
      operationId: createKnowledgeGoldenSet
      parameters:
        - name: Idempotency-Key
          in: header
          required: false
          description: >-
            1–255 printable ASCII characters. A repeat with the same key and
            request never runs the write again, a 5xx included. It returns the
            first response (`Idempotency-Replayed: true`) only when the route's
            current authorization policy permits disclosure; otherwise `409`
            `idempotency_result_unavailable`. Honoured for External App JWTs and
            platform and MCP keys; an External App's static companion key is not
            protected.
          schema:
            type: string
            minLength: 1
            maxLength: 255
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/CreateKnowledgeGoldenSetRequest'
      responses:
        '201':
          description: The golden set created
          content:
            application/json:
              schema:
                type: object
                properties:
                  data:
                    $ref: '#/components/schemas/KnowledgeGoldenSet'
                  meta:
                    $ref: '#/components/schemas/PlatformMeta'
                required:
                  - data
                  - meta
          headers:
            X-Request-Id:
              $ref: '#/components/headers/RequestId'
            RateLimit-Limit:
              $ref: '#/components/headers/RateLimitLimit'
            RateLimit-Remaining:
              $ref: '#/components/headers/RateLimitRemaining'
            RateLimit-Reset:
              $ref: '#/components/headers/RateLimitReset'
        '400':
          description: >-
            Bad request: an unknown field; a blank name or scope; 0 or over 50
            namespaces; 0 or over 100 cases, a repeated or blank `caseId`, a
            blank query, or 0 or over 50 expected keys; an unknown status; or
            `idempotency_key_invalid`
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ErrorResponse400Keyed'
          headers:
            X-Request-Id:
              $ref: '#/components/headers/RequestId'
            RateLimit-Limit:
              $ref: '#/components/headers/RateLimitLimit'
            RateLimit-Remaining:
              $ref: '#/components/headers/RateLimitRemaining'
            RateLimit-Reset:
              $ref: '#/components/headers/RateLimitReset'
        '401':
          description: Unauthorized
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ErrorResponse401'
          headers:
            X-Request-Id:
              $ref: '#/components/headers/RequestId'
            RateLimit-Limit:
              $ref: '#/components/headers/RateLimitLimit'
            RateLimit-Remaining:
              $ref: '#/components/headers/RateLimitRemaining'
            RateLimit-Reset:
              $ref: '#/components/headers/RateLimitReset'
        '403':
          description: >-
            `forbidden`: the credential lacks the required scope, or a namespace
            in the body is outside the caller's readable namespaces (recorded,
            naming no `requiredPermission`)
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ErrorResponse403'
          headers:
            X-Request-Id:
              $ref: '#/components/headers/RequestId'
            RateLimit-Limit:
              $ref: '#/components/headers/RateLimitLimit'
            RateLimit-Remaining:
              $ref: '#/components/headers/RateLimitRemaining'
            RateLimit-Reset:
              $ref: '#/components/headers/RateLimitReset'
        '409':
          description: >-
            `idempotency_key_in_flight` (the first request with this key is
            still running), `idempotency_outcome_unknown` (it stopped without
            recording an outcome; reconcile the resource, then use a new key
            only if another effect is appropriate) or
            `idempotency_result_unavailable` (it completed, but its response
            cannot be disclosed; reconcile the resource before using a new key)
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ErrorResponse409Keyed'
          headers:
            X-Request-Id:
              $ref: '#/components/headers/RequestId'
            RateLimit-Limit:
              $ref: '#/components/headers/RateLimitLimit'
            RateLimit-Remaining:
              $ref: '#/components/headers/RateLimitRemaining'
            RateLimit-Reset:
              $ref: '#/components/headers/RateLimitReset'
        '422':
          description: >-
            `idempotency_key_reuse`: the key was already used for a different
            request
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ErrorResponse422Keyed'
          headers:
            X-Request-Id:
              $ref: '#/components/headers/RequestId'
            RateLimit-Limit:
              $ref: '#/components/headers/RateLimitLimit'
            RateLimit-Remaining:
              $ref: '#/components/headers/RateLimitRemaining'
            RateLimit-Reset:
              $ref: '#/components/headers/RateLimitReset'
        '429':
          $ref: '#/components/responses/RateLimited'
        '500':
          description: >-
            `internal_error`: the request failed (an embeddings or model
            provider error, for example). The body names only the `requestId`.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ErrorResponse500'
          headers:
            X-Request-Id:
              $ref: '#/components/headers/RequestId'
            RateLimit-Limit:
              $ref: '#/components/headers/RateLimitLimit'
            RateLimit-Remaining:
              $ref: '#/components/headers/RateLimitRemaining'
            RateLimit-Reset:
              $ref: '#/components/headers/RateLimitReset'
        '503':
          $ref: '#/components/responses/LimiterUnavailable'
      security:
        - BearerAuth: []
        - PlatformApiKey: []
        - PersonalKey: []
components:
  schemas:
    CreateKnowledgeGoldenSetRequest:
      type: object
      properties:
        name:
          type: string
          minLength: 1
        scope:
          type: string
          minLength: 1
        namespaces:
          type: array
          items:
            type: string
            minLength: 1
          minItems: 1
          maxItems: 50
        cases:
          type: array
          items:
            type: object
            properties:
              caseId:
                type: string
                minLength: 1
              query:
                type: string
                minLength: 1
              expectedKeys:
                type: array
                items:
                  type: string
                  minLength: 1
                minItems: 1
                maxItems: 50
            required:
              - caseId
              - query
              - expectedKeys
            additionalProperties: false
          minItems: 1
          maxItems: 100
        status:
          type: string
          enum:
            - active
            - archived
      required:
        - name
        - scope
        - namespaces
        - cases
      additionalProperties: false
      description: >-
        A new golden set: its name, the scope it measures, the namespaces to
        search (1–50, each one the caller may read) and its cases. Each case is
        a query and the golden keys a good retrieval returns for it (a case key,
        or a synced entry's source id), 1–50 per case; 1–100 cases with distinct
        `caseId`s. `status` defaults to `active`.
    KnowledgeGoldenSet:
      type: object
      properties:
        id:
          type: string
          minLength: 1
        name:
          type: string
        scope:
          type: string
        namespaces:
          type: array
          items:
            type: string
        cases:
          type: array
          items:
            type: object
            properties:
              caseId:
                type: string
                minLength: 1
              query:
                type: string
                minLength: 1
              expectedKeys:
                type: array
                items:
                  type: string
                  minLength: 1
                minItems: 1
                maxItems: 50
            required:
              - caseId
              - query
              - expectedKeys
            additionalProperties: false
        status:
          type: string
          enum:
            - active
            - archived
        createdAt:
          type: number
        lastRunRequestedAt:
          type:
            - number
            - 'null'
      required:
        - id
        - name
        - scope
        - namespaces
        - cases
        - status
        - createdAt
        - lastRunRequestedAt
      additionalProperties: false
      description: >-
        One retrieval golden set whose every namespace the caller may read: the
        golden cases a run measures and the namespaces each is searched in. Each
        case is a query and the golden keys a good retrieval returns for it (a
        case key, or a synced entry's source id), 1–50 per case; 1–100 cases
        with distinct `caseId`s. `lastRunRequestedAt` is when a run was last
        requested on demand (null if never). Never names who created it.
    PlatformMeta:
      type: object
      properties:
        requestId:
          type: string
        hasMore:
          type: boolean
        nextCursor:
          type:
            - string
            - 'null'
        asOf:
          anyOf:
            - type: integer
            - type: 'null'
      required:
        - requestId
      additionalProperties: true
    ErrorResponse400Keyed:
      allOf:
        - $ref: '#/components/schemas/ErrorResponse'
        - type: object
          properties:
            error:
              type: object
              properties:
                code:
                  type: string
                  enum:
                    - bad_request
                    - allowed_scopes_removed
                    - idempotency_key_invalid
    ErrorResponse401:
      allOf:
        - $ref: '#/components/schemas/ErrorResponse'
        - type: object
          properties:
            error:
              type: object
              properties:
                code:
                  type: string
                  enum:
                    - unauthorized
    ErrorResponse403:
      allOf:
        - $ref: '#/components/schemas/ErrorResponse'
        - type: object
          properties:
            error:
              type: object
              properties:
                code:
                  type: string
                  enum:
                    - forbidden
                    - permission_denied
                    - operator_stopped
    ErrorResponse409Keyed:
      allOf:
        - $ref: '#/components/schemas/ErrorResponse'
        - type: object
          properties:
            error:
              type: object
              properties:
                code:
                  type: string
                  enum:
                    - conflict
                    - invalid_state
                    - idempotency_key_in_flight
                    - idempotency_result_unavailable
                    - idempotency_outcome_unknown
    ErrorResponse422Keyed:
      allOf:
        - $ref: '#/components/schemas/ErrorResponse'
        - type: object
          properties:
            error:
              type: object
              properties:
                code:
                  type: string
                  enum:
                    - unprocessable
                    - idempotency_key_reuse
    ErrorResponse500:
      allOf:
        - $ref: '#/components/schemas/ErrorResponse'
        - type: object
          properties:
            error:
              type: object
              properties:
                code:
                  type: string
                  enum:
                    - internal_error
    ErrorResponse:
      type: object
      properties:
        error:
          type: object
          properties:
            code:
              type: string
              enum:
                - unauthorized
                - forbidden
                - permission_denied
                - not_found
                - resource_missing
                - method_not_allowed
                - bad_request
                - unprocessable
                - conflict
                - invalid_state
                - gone
                - payment_required
                - idempotency_key_invalid
                - idempotency_key_in_flight
                - idempotency_key_reuse
                - idempotency_result_unavailable
                - idempotency_outcome_unknown
                - allowed_scopes_removed
                - governance_door_stopped
                - operator_stopped
                - payload_too_large
                - rate_limited
                - internal_error
            message:
              type: string
            docUrl:
              type: string
              format: uri
            reason:
              type: string
              enum:
                - requester_unavailable
                - invalid_body
                - form_invalid
                - form_secret_like
                - eligibility_invalid
                - due_out_of_range
                - no_candidates
                - org_unit_required
                - requester_open_cap
                - person_open_cap
                - rate_limited
                - role_giving_refused
            requiredPermission:
              type:
                - string
                - 'null'
            personRequired:
              type: boolean
            decisionId:
              type: string
            capHeadroom:
              type: object
              properties:
                remainingCents:
                  type: number
                capCents:
                  type: number
                consumedCents:
                  type: number
                estimateCents:
                  type: number
              required:
                - remainingCents
                - capCents
                - consumedCents
                - estimateCents
              additionalProperties: false
            details:
              type: array
              items:
                type: object
                properties:
                  path:
                    type: string
                  param:
                    type: string
                  message:
                    type: string
                  code:
                    type: string
                    enum:
                      - required
                      - invalid_type
                      - invalid_value
                      - invalid_format
                      - too_small
                      - too_big
                      - unknown_field
                      - invalid_json
                      - invalid
                required:
                  - path
                  - param
                  - message
                  - code
                additionalProperties: false
              maxItems: 50
          required:
            - code
            - message
          additionalProperties: true
        meta:
          type: object
          properties:
            requestId:
              type: string
            hasMore:
              type: boolean
            nextCursor:
              type:
                - string
                - 'null'
            asOf:
              anyOf:
                - type: integer
                - type: 'null'
          required:
            - requestId
          additionalProperties: true
      required:
        - error
        - meta
      additionalProperties: true
  headers:
    RequestId:
      description: >-
        This answer's request id, the same as the envelope's `meta.requestId`:
        `req_` and 16 lowercase hex digits. Every response carries it. Quote it
        to support, or look the request up with `GET
        /api/v3/me/requests/{requestId}`.
      schema:
        type: string
    RateLimitLimit:
      description: >-
        The capacity of the bucket this request spent from: the most requests it
        admits at once. Sent on every response to a request that spent from its
        bucket, whatever its status: a 401 or 503 the operation answers after
        that spend carries it too. Absent on a refusal before the spend (a 401
        from authentication, a stopped `api_keys` door's 503
        `governance_door_stopped`, a credential's 403 at sign-in), on a route's
        own tighter 429, and on the limiter's own 503, which could not spend.
      schema:
        type: integer
        minimum: 0
    RateLimitRemaining:
      description: >-
        Requests left in the bucket this request spent from; 0 on a refusal.
        While the limiter is degraded (its fallback bucket admitted the
        request), an estimate.
      schema:
        type: integer
        minimum: 0
    RateLimitReset:
      description: Seconds until the bucket this request spent from is full again.
      schema:
        type: integer
        minimum: 0
    RetryAfter:
      description: >-
        Seconds to wait before sending the request again. Sent on a 429 and on a
        transient 503.
      schema:
        type: integer
        minimum: 1
  responses:
    RateLimited:
      description: >-
        `rate_limited`: this credential, or the organization as a whole, has
        spent the requests of this operation's class for now. Each credential
        has three buckets: `read` (GET, HEAD), `write` (every other method) and
        `expensive` (an operation that starts a model turn, a run, a billable VM
        or a bulk job); each operation names its class in `x-forge-rate-limit`
        and spends from that bucket only. Every credential's requests also spend
        from the organization ceiling, one bucket per class at ten times a
        credential's limit, shared by every credential of the deployment; the
        message names which one refused. Wait `Retry-After` seconds, then retry.
        Every authenticated response carries the `RateLimit-*` headers for the
        bucket it spent from, so a client can slow down before it is refused;
        `GET /api/v2/me/limits` answers all three buckets and the ceiling.
      headers:
        Retry-After:
          $ref: '#/components/headers/RetryAfter'
        RateLimit-Limit:
          $ref: '#/components/headers/RateLimitLimit'
        RateLimit-Remaining:
          $ref: '#/components/headers/RateLimitRemaining'
        RateLimit-Reset:
          $ref: '#/components/headers/RateLimitReset'
        X-Request-Id:
          $ref: '#/components/headers/RequestId'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
    LimiterUnavailable:
      description: >-
        `internal_error`: The server could not complete the request. Wait for
        `Retry-After`, then retry; resend a write only with the same
        `Idempotency-Key`. Forge answers it when its rate limiter is briefly
        unavailable (a read or write request falls back to a second bucket
        before it is refused; an `expensive` request is refused at once), or
        when a transient runtime fault stops the request.
      headers:
        Retry-After:
          $ref: '#/components/headers/RetryAfter'
        RateLimit-Limit:
          $ref: '#/components/headers/RateLimitLimit'
        RateLimit-Remaining:
          $ref: '#/components/headers/RateLimitRemaining'
        RateLimit-Reset:
          $ref: '#/components/headers/RateLimitReset'
        X-Request-Id:
          $ref: '#/components/headers/RequestId'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
  securitySchemes:
    BearerAuth:
      type: http
      scheme: bearer
      bearerFormat: JWT
      description: >-
        An External App: the JWT it obtains from `POST /oauth/token` with OAuth2
        client_credentials. Pass as `Authorization: Bearer <jwt>`. An app's
        static companion key also authenticates as the app, but is deprecated
        and is not protected by `Idempotency-Key`.
    PlatformApiKey:
      type: http
      scheme: bearer
      description: >-
        A platform API key, created in the Console under Developers. Pass as
        `Authorization: Bearer <key>`. It acts as itself, under its role grants
        narrowed by its scopes.
    PersonalKey:
      type: http
      scheme: bearer
      description: >-
        A Personal key (`fmcp_…`), bound to the person who created it, or an
        interactive MCP sign-in that resolves to one. Pass as `Authorization:
        Bearer <key>`. It acts for that person, under their role grants narrowed
        by its scopes; an operation that acts for a person admits only this
        kind.

````

This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.