> ## Documentation Index > Fetch the complete documentation index at: https://formbricks.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Search feedback records semantically > Embeds `query` and returns the ids of the records in the dataset whose text is closest to it, with cosine similarity scores. Only records with a non-empty `value_text` are embedded, so only those can match. The gateway reads `tenant_id` from the JSON body to authorize the call. Requires an API key with **read** permission (or higher) on a workspace the dataset is assigned to. Results carry ids and the embedded text only — fetch a match with `GET /api/v3/feedbackRecords/{id}` for the full record. Answers **503** on instances without an embedding model configured. ## OpenAPI ````yaml /api-v3-reference/openapi.yml post /api/v3/feedbackRecords/search/semantic openapi: 3.1.1 info: title: Formbricks API v3 description: > **Surveys**: **GET /api/v3/surveys**, **POST /api/v3/surveys**, **POST /api/v3/surveys/generate**, **POST /api/v3/surveys/validate**, **GET /api/v3/surveys/{surveyId}**, **PATCH /api/v3/surveys/{surveyId}**, **PATCH /api/v3/surveys/{surveyId}/blocks**, **PUT /api/v3/surveys/{surveyId}/blocks/order**, and **DELETE /api/v3/surveys/{surveyId}**. **Custom CSS**: survey CSS through `customCss` on **POST /api/v3/surveys**, **GET** and **PATCH /api/v3/surveys/{surveyId}**, workspace CSS through **GET** and **PATCH /api/v3/workspaces/{workspaceId}/custom-css**, and previews through `operation: customCss` on **POST /api/v3/surveys/validate**. Writes take and reads return source only; the server compiles it, returns removed constructs as top-level `warnings`, and never accepts compiled output. **Responses**: **GET /api/v3/responses**, **GET /api/v3/responses/count**, **POST /api/v3/responses/validate**, **POST /api/v3/responses/batch-delete**, **GET /api/v3/responses/{responseId}**, **POST /api/v3/responses**, **PATCH /api/v3/responses/{responseId}**, and **DELETE /api/v3/responses/{responseId}**. **Workflows extension**: **GET /api/v3/workflows**, **POST /api/v3/workflows**, **GET /api/v3/workflows/{workflowId}**, **PATCH /api/v3/workflows/{workflowId}**, **DELETE /api/v3/workflows/{workflowId}**, **POST /api/v3/workflows/{workflowId}/duplicate**, **POST /api/v3/workflows/{workflowId}/enable**, **POST /api/v3/workflows/{workflowId}/disable**, **POST /api/v3/workflows/{workflowId}/archive**, **POST /api/v3/workflows/{workflowId}/unarchive**, **POST /api/v3/workflows/{workflowId}/test**, **GET /api/v3/workflows/runs**, and **GET /api/v3/workflows/runs/{runId}**. Authenticate all management endpoints with **session cookie** or **`x-api-key`** (management key with access to the workspace). **Spec location:** source of truth is the multi-file tree at `docs/api-v3-reference/src/` (root `openapi.yml` plus one file per path and component); `docs/api-v3-reference/openapi.yml` is the generated single-file bundle (alongside v2 at `docs/api-v2-reference/openapi.yml`). **workspaceId** `workspaceId` is the canonical container identifier for this API. It is a **required query parameter on collection read endpoints** (`GET /api/v3/surveys`, `GET /api/v3/action-classes`, `GET /api/v3/contact-attribute-keys`) because those operate within a workspace. **`POST /api/v3/surveys` instead requires `workspaceId` in the request body**, not as a query parameter. Single-resource endpoints (`GET`/`PATCH`/`DELETE /api/v3/surveys/{surveyId}`) locate the resource by its globally-unique id and resolve the workspace from it, so they do **not** accept a `workspaceId` query parameter — sending one returns **400** with an `invalid_params` entry naming the key itself (`name: "workspaceId"`, `reason: "Unsupported field 'workspaceId'"`), rather than one naming the body. **Auth** Authenticate with either a session cookie or **`x-api-key`**. In dual-auth mode, V3 checks the API key first when the header is present, otherwise it uses the session path. Unauthenticated callers get **401** before query validation. **Pagination** Cursor-based pagination with **limit** + opaque **cursor** token. Responses return `meta.nextCursor`; pass that value back as `cursor` to fetch the next page. Survey list responses also include `meta.totalCount`, the total number of surveys matching the current filters across all pages. Workflow list responses do not require total counts in Scope 1. There is no `offset` in this contract. The page-size **maximum is 250** for responses, surveys, action classes and contact attribute keys; only the *default* varies by resource (20 for surveys, 50 for the workspace reference lists). **The workflows and workflow-run lists remain at 100** — their bound lives in the `packages/workflows` contracts and moves with that package, so each documents the value it actually enforces. A `limit` above an endpoint's maximum is rejected with **400** rather than silently reduced, consistent with v3 rejecting unrecognized query parameters instead of ignoring them. Callers should not infer the end of a collection from a page shorter than `limit`; `meta.nextCursor` of `null` is the only end-of-collection signal. Note that array-valued *input* bounds are a separate concern and stay at 100 — `filter[id][in]` and the batch-delete id list are bounded by write and payload risk, not by page cost. **Strictness** A query parameter that an operation does not declare returns **400**, with the offending key reported in `invalid_params[].name` — including on operations that declare no query parameters at all. Unknown parameters are rejected rather than ignored on purpose: a silently-dropped filter returns *more* data than the caller believes they asked for, and a typo in a filter name is indistinguishable from no filter at all. The same applies to request bodies, where unsupported fields are rejected instead of stripped. Unrecognized **headers** are tolerated, as gateways and proxies legitimately add them; strictness is about parameters and body fields, not transport metadata. This is the rule for the surveys, responses, tags and workspace reference-list operations. The workflows family does not enforce it today, so treat parameter strictness there as unspecified rather than guaranteed. **Request limits** A request body may be at most **2 MB**; a larger one is refused with **413** before it is read. Inside a body, **no array may hold more than 1 000 items, and all arrays together may hold at most 50 000 elements**, at any depth — a body that exceeds either bound is refused with **400** and a single `invalid_params` entry naming the offending array, before any field is validated. The same holds for nesting: a body may nest at most **256** levels of objects and arrays, counted from the root, and a deeper one is refused the same way, naming where it went too deep. Fields with a tighter bound document it as `maxItems` on the field (a survey's `blocks`, `endings`, `variables`, `languages`, `hiddenFields.fieldIds` and `distribution.triggers` since 0.2.0). A 400 or 422 lists at most **50** `invalid_params` entries; when there are more, the last entry says how many were left out. Since 0.2.0. **Filtering** Filters use explicit operator-style query parameters under the **`filter[...]` family**. The survey list supports `filter[name][contains]`, `filter[status][in]`, and `filter[type][in]`. The workflow list supports `filter[name][contains]` and `filter[status][in]`; the workflow run list supports `filter[status][in]` and `filter[isDryRun][eq]`. Resource-scoping identifiers (`workspaceId`, `workflowId`, `responseId`) stay plain query parameters. Multi-value filters use repeated keys or comma-separated values (e.g. `filter[status][in]=draft&filter[status][in]=inProgress` or `filter[status][in]=draft,inProgress`). Sorting remains a flat `sortBy` query parameter. **Responses** The response resource is **self-describing**. Answers come back as a typed `answers[]` array in which each entry carries the element's id, label, type and value, rather than as a map keyed by element id — so a consumer holding only a response can read it without fetching the survey and joining. Hidden fields, survey variables and auto-captured context are one `embeddedData[]` collection rather than three parallel concepts, each entry declaring a `kind` and a `type`. Data whose element has since been deleted is reported in `unresolved[]` instead of being dropped silently. Labels are resolved at read time against the current survey definition, and the `resolution` object discloses that policy along with the survey's `updatedAt` so a consumer can detect wording drift. Writes take answers in the **stored map form** (`data`) while reads return the typed array. Resolving ids into labelled answers is unambiguous; the reverse would mean matching caller-supplied labels back onto option ids, so the write side keeps the stored shape. To stop that costing round-trippability, the **detailed read returns `data` unchanged** — so a single-response read can be edited and sent back as `PATCH { "data": … }`, and the typed `answers[]` array is an addition rather than a replacement. One case where that round trip does not close: a write bound applies to `data` on the way in (`ResponseDataMapInput`) and none applies to the stored map, because v1 and v2 wrote it without one. A response carrying more answer keys than the write accepts therefore cannot be corrected through `PATCH` at all — the whole map is rejected, and sending a smaller one is not a workaround, because `data` is replaced wholesale and every key left out is deleted. Correct those through the v1/v2 management API or the dashboard. This is reachable only on surveys with more elements than the bound allows. Embedded Data is the exception, and it is symmetric rather than round-tripped: hidden fields and variables are read and written through `embeddedData`, addressed by **name** in both directions. No raw map of them is echoed, because a name is already what a write accepts — publishing the storage keys underneath (a variable's is its cuid) would expose a layout the caller has no use for. `data` therefore carries element answers only. Of those two, only `data` carries no `readOnly`, because it is the same map on the way in and on the way out. Embedded Data is writable through the **request's own** `embeddedData` map, not through the response field of that name — so `embeddedData[]`, like every other derived view on the resource, is marked `readOnly`, per its meaning of server-managed. `ResponseResource` lists them all. The list view omits `data`, since a bulk reader is not editing. `POST /api/v3/responses/validate` exists so a caller can see what a map would produce, and what a write would do, before committing to it. `createdAt` and `updatedAt` are **server-owned** on this resource: they cannot be supplied on create or changed by a patch. Response metering counts rows by `(surveyId, createdAt)`, and a client-chosen `createdAt` also lets rows be inserted into the middle of another consumer's already-paginated stream. Backdating stays available on v2 until a dedicated import path exists. The respondent's IP address and the `contactAttributes` snapshot that v1/v2 expose are not part of this resource in any view; that snapshot is copied from the contact at submission time and never refreshed, so it is stale PII by construction. Response filtering is allow-listed and index-backed: `filter[createdAt][gte|gt|lte|lt]` (at most one lower and one upper bound — `gte` xor `gt`, `lte` xor `lt`), `filter[finished][eq]`, `filter[language][in]` and `filter[id][in]`. `filter[finished][eq]` and `filter[language][in]` require `surveyId` or `contactId` as well, because both are heap residuals bounded only by an anchored index range. Per-answer, hidden-field, submission-metadata, contact-attribute, quota and tag filters are **not** offered; each is a JSON path or an unindexed join on the largest table in the product. The in-app dashboard does offer them, so this API is knowingly behind the dashboard here — closing the gap requires index work and is tracked separately rather than quietly omitted. `GET /api/v3/responses/count` answers "how many match?" directly. The list endpoint's `meta.totalCount` covers a caller that is already paging and is **opt-in** via `includeTotalCount` (default `false`, unlike the survey list) because responses are the highest-cardinality table in the product. Both response counts are **capped at 10 000** and carry an exactness flag — `meta.totalCountRelation` on the list, `data.relation` on `/count` — reading `eq` when the figure is exact and `gte` when the cap was reached and it is a lower bound. An exact count over this table grows superlinearly with the data while a capped one stays flat, so an uncapped total would be the only part of a response whose cost is unbounded. `/count` accepts `precision=exact` for the cases that genuinely need a precise figure, documented as the deliberate slow path. The survey list's `totalCount` stays exact and uncapped: surveys are low-cardinality and that count has a covering index built for it. Bulk deletion is `POST /api/v3/responses/batch-delete`, a custom method rather than a collection-level `DELETE`, so a call that loses its filters cannot degrade into erasing everything in scope. It scope-filters rather than rejecting: foreign and already-deleted ids are ignored and the call reports how many rows it removed, which keeps it both leak-free and safe to retry. **Security** Missing/forbidden workspace returns **403** with a generic message (not **404**) so resource existence is not leaked. Workflow and workflow-run detail endpoints also return **403** for unknown IDs or inaccessible IDs to avoid IDOR leaks. List responses use `private, no-store`. **AI survey creation** `POST /api/v3/surveys/generate` returns a draft `POST /api/v3/surveys` payload plus validation metadata. It does not create a survey. It generates survey content for both `link` and `app` surveys; for `app` it seeds a default `distribution` (display once, no triggers, no targeting) that you finish configuring before publishing. Prompt text is sent to the configured AI provider for generation, but it is not stored by this endpoint, not logged by default, and not included in audit data. **Feedback Records** `/api/v3/feedbackRecords` is served by the **API gateway** (Envoy on Formbricks Cloud, Traefik in the one-click Docker install), which authorizes each call against the dataset and forwards it to the feedback store — not by the application routes that serve the rest of v3. Three consequences: the payload is the store's own, so members and filters are **`snake_case`** (`tenant_id`, `value_text`) rather than v3's camelCase; the dataset is named **`tenant_id`** on the wire and is required on every collection operation (it is the feedback dataset id, which the MCP `list_feedback_datasets` tool reports as `dataset_id`); and errors come in two shapes — gateway refusals (**400**, **401**, **403**, **413**, **503**) are `text/plain`, while the store's own problems are `application/problem+json` with a stable `code`. These eight operations are the whole of the gateway-served surface; any other path it receives under the prefix answers **400**. Authenticate with **`x-api-key`**: the key needs a permission on a workspace the dataset is assigned to — `read` to list, get and search, `write` to create and update, `manage` to delete — and may update or delete only in a dataset assigned to exactly one workspace. **OpenAPI** This spec is **not** produced by `pnpm generate-api-specs` (that script only builds v2 → `docs/api-v2-reference/openapi.yml`). When the route contract changes, edit the files under `docs/api-v3-reference/src/`, then run `pnpm api:v3:bundle` to regenerate the committed single-file bundle. `pnpm api:v3:lint` lints the source tree; `pnpm api:v3:check` (CI) fails when the bundle is stale. **Workflows Scope 1** Workflows are a follow-up extension of the existing v3 Survey API work. The Survey API establishes API-first survey authoring; Workflows build on that same v3 style to automate actions around survey responses through workspace-scoped JSON definitions. Scope 1 models the existing Follow-ups capability as workflows: one `response.completed` trigger with an optional `endingCardIds` filter, and `send_email` actions with Follow-up email field parity. The definition and run shapes mirror the shared Zod schemas in `packages/workflows` (ENG-1100), which are the implementation source of truth for this contract. Workflows use `draft`, `enabled`, and `disabled` lifecycle states plus an `archived` soft-delete state. Status only changes through the lifecycle endpoints (`enable`, `disable`, `archive`, `unarchive`); it is not writable via `POST` or `PATCH`. Enabling validates that the definition is executable and snapshots an immutable workflow version; runs reference that snapshot via `workflowVersionId`. `POST .../test` is a dry run: it validates the definition and resolves the trigger's references synchronously, creating no run and sending no email, and returns `{ workflowId, ok, problems }` directly. Real runs execute asynchronously and are polled via `GET /api/v3/workflows/runs/{runId}`. **Overview migration note** The v3-backed survey overview page intentionally removes actions that are not yet exposed by this contract: `Created by` filtering, `Duplicate`, `Copy...`, `Preview`, and `Copy link`. **Next steps (out of scope for this spec)** Optional ETag/304, field selection, translation-specific writes, and survey version history. version: 0.2.0 x-implementation-notes: route: apps/web/app/api/v3/surveys/route.ts generate-route: apps/web/app/api/v3/surveys/generate/route.ts query-parser: apps/web/app/api/v3/surveys/parse-v3-surveys-list-query.ts auth: apps/web/app/api/v3/lib/auth.ts workspace-resolution: apps/web/app/api/v3/lib/workspace-context.ts workflow-contract-ticket: ENG-1101 workflow-shared-schemas: packages/workflows/src/types (feat/workflows-types) workflow-prisma-schema: packages/database/schema/workflows.prisma (feat/workflows-prisma-schema) openapi-generated: false spec-source: docs/api-v3-reference/src (regenerate bundle with `pnpm api:v3:bundle`) responses-contract-ticket: ENG-2868 responses-rfc: v3 Responses API RFC sections 2b (resource) and 2c (endpoints) responses-stored-shape: packages/types/responses.ts (ZResponse, ZResponseData, ZResponseInput) responses-prisma-model: packages/database/schema/main.prisma (model Response) pagination-model: cursor cursor-pagination: supported servers: - url: https://app.formbricks.com description: >- Formbricks Cloud. Paths in this spec already include the `/api/v3` prefix; self-hosted instances substitute their own origin. security: [] paths: /api/v3/feedbackRecords/search/semantic: post: tags: - V3 Feedback Records summary: Search feedback records semantically description: > Embeds `query` and returns the ids of the records in the dataset whose text is closest to it, with cosine similarity scores. Only records with a non-empty `value_text` are embedded, so only those can match. The gateway reads `tenant_id` from the JSON body to authorize the call. Requires an API key with **read** permission (or higher) on a workspace the dataset is assigned to. Results carry ids and the embedded text only — fetch a match with `GET /api/v3/feedbackRecords/{id}` for the full record. Answers **503** on instances without an embedding model configured. operationId: searchFeedbackRecordsV3 parameters: - $ref: '#/components/parameters/SimilarityLimitQuery' - $ref: '#/components/parameters/SimilarityCursorQuery' - $ref: '#/components/parameters/MinScoreQuery' requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/FeedbackRecordSemanticSearchRequest' example: query: complaints about waiting times at the gates tenant_id: ph7zv3w2u1x5k9d8m4q6c0bn responses: '200': description: Matches retrieved content: application/json: schema: $ref: '#/components/schemas/FeedbackRecordSimilarityResult' examples: matches: summary: Two matches above the threshold value: data: - feedback_record_id: 018e1234-5678-7abc-8ef0-123456789abd score: 0.71 field_label: What could we improve? value_text: Queues at the north gate were far too long. - feedback_record_id: 018e1234-5678-7abc-8ef0-1234567890aa score: 0.64 field_label: Anything else? value_text: >- Took 40 minutes to get in. Please open more turnstiles. limit: 10 '400': $ref: '#/components/responses/FeedbackRecordsBadRequest' '401': $ref: '#/components/responses/FeedbackRecordsUnauthorized' '403': $ref: '#/components/responses/FeedbackRecordsForbidden' '413': $ref: '#/components/responses/FeedbackRecordsPayloadTooLarge' '500': $ref: '#/components/responses/FeedbackRecordsInternalServerError' '503': $ref: '#/components/responses/FeedbackRecordsServiceUnavailable' security: - apiKeyAuth: [] components: parameters: SimilarityLimitQuery: in: query name: limit schema: type: integer default: 10 description: >- Page size, at most 100. Unlike the `limit` on `GET /api/v3/feedbackRecords`, this one is never rejected: the search endpoints **clamp** instead of answering 400. A value above 100 is reduced to 100, and `0`, a negative number or a non-numeric value falls back to the default of 10. Read the `limit` in the response to see what was actually applied. SimilarityCursorQuery: in: query name: cursor schema: type: string description: >- Opaque keyset cursor returned as `next_cursor` by the previous page. Omit on the first request. A malformed cursor answers **400**. MinScoreQuery: in: query name: min_score schema: type: number format: float default: 0.7 description: >- Only matches with `score >= min_score` are returned. Like `limit`, this value is **clamped rather than rejected**: above `1` it becomes `1`, below `0` it becomes `0`, and a non-numeric value falls back to the default of `0.7`. That default is conservative — on the embedding model we measured, query→record similarities for genuinely relevant text often landed around `0.55–0.7` — so if a good query returns nothing, lower it (e.g. `0.5`) and filter on the returned `value_text`. Score distributions differ per model, so treat those numbers as an observation rather than a guarantee. schemas: FeedbackRecordSemanticSearchRequest: type: object additionalProperties: false required: - query - tenant_id properties: query: type: string minLength: 1 description: >- Natural-language search text. It is embedded and compared to record embeddings by cosine similarity. tenant_id: type: string format: cuid2 description: >- The feedback dataset to search (the dataset id). Results never cross datasets. FeedbackRecordSimilarityResult: type: object additionalProperties: false required: - data - limit properties: data: type: array maxItems: 100 items: $ref: '#/components/schemas/FeedbackRecordSimilarityItem' description: Matches ordered by descending `score`. limit: type: integer description: The page size that was applied. next_cursor: type: string description: >- Opaque keyset cursor for the next page. Present only when a full page was returned and there may be more results. Pass it back unchanged as `cursor`. FeedbackRecordSimilarityItem: type: object additionalProperties: false required: - feedback_record_id - score - field_label - value_text properties: feedback_record_id: type: string format: uuid description: >- The matching record. Fetch it with `GET /api/v3/feedbackRecords/{id}` for the full payload. score: type: number format: double minimum: 0 maximum: 1 description: Cosine similarity, `0` unrelated to `1` identical. field_label: type: string description: >- The record's question text (part of what was embedded). Empty when the record has no label. value_text: type: string description: >- The record's text answer (the text that was embedded). Empty when it has since been cleared — only records with a non-empty `value_text` are embedded in the first place. HubProblem: type: object additionalProperties: false description: >- RFC 9457 problem details (`application/problem+json`) as produced by the feedback store behind the API gateway. Branch on the stable `code` member rather than on `title` or `type`. Note the member names differ from the rest of v3 (`request_id`, not `requestId`) — the payload is passed through unchanged. required: - title - status - code - request_id properties: type: type: string format: uri description: A URI reference identifying the problem type. title: type: string description: >- Short, human-readable summary of the problem type; stable across occurrences. status: type: integer description: HTTP status code. code: type: string description: >- Stable, machine-readable error code. `service_unavailable` means a feature or dependency is temporarily unavailable and the request may be retried later; `tenant_write_conflict` means the write collided with an in-progress dataset purge and may be retried unchanged. Every other code is terminal until the request itself changes. enum: - validation - bad_request - unauthorized - forbidden - not_found - conflict - tenant_write_conflict - method_not_allowed - content_too_large - service_unavailable - internal_server_error detail: type: string description: Human-readable explanation specific to this occurrence. instance: type: string description: >- URI reference identifying this occurrence — the request path **as the feedback store saw it** (`/v1/feedback-records/...`), not the `/api/v3/feedbackRecords/...` path you called. request_id: type: string description: >- Identifier for this request, also returned in the `X-Request-ID` response header; include it when reporting a problem. details: type: object additionalProperties: true description: >- Optional structured context, e.g. `resource_type` on a not-found problem. invalid_params: type: array maxItems: 100 items: $ref: '#/components/schemas/HubInvalidParam' description: Field-level validation failures. Present on `validation` problems. HubInvalidParam: type: object additionalProperties: false description: >- A single invalid request parameter (RFC 9457 `invalid_params` extension member). required: - name - reason properties: name: type: string description: >- Dotted path to the offending request field, e.g. `field_type` or `items[0].field_type`. reason: type: string description: >- Human- and agent-readable explanation of why the value is invalid, naming allowed values or constraints so the request can be corrected without guessing. responses: FeedbackRecordsBadRequest: description: >- Bad Request. From the **gateway** (`text/plain`): a missing or malformed `tenant_id`; a request body that is not a single JSON object; a body whose top-level keys can be read more than one way (a repeated key, a key with non-ASCII characters, or a key that differs from `tenant_id` only in letter case); a `tenant_id` query parameter given more than once; or a path under `/api/v3/feedbackRecords` that is not one of the documented operations. A record id that is not a UUID and a path such as `/count` both fall in that last case and share the one `Unsupported FeedbackRecords route` body, so the two cannot be told apart from the response. From the **feedback store** (`application/problem+json`): schema validation failures — an unknown enum value, an inverted range, a malformed or re-ordered cursor, or a string with NULL bytes or invalid UTF-8. content: text/plain: schema: type: string examples: missingTenant: summary: Gateway — tenant_id missing or malformed value: Invalid or missing tenant_id invalidBody: summary: >- Gateway — body is not a single JSON object, or has a repeated or non-ASCII key value: Invalid request body ambiguousTenant: summary: >- Gateway — tenant_id given in more than one spelling, or repeated in the query value: Ambiguous tenant_id unsupportedRoute: summary: Gateway — not a documented operation value: Unsupported FeedbackRecords route application/problem+json: schema: $ref: '#/components/schemas/HubProblem' examples: invertedRange: summary: Feedback store — a range filter supplied backwards value: type: https://hub.formbricks.com/problems/validation title: Validation Failed status: 400 code: validation detail: One or more request parameters are invalid request_id: 0190a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b invalid_params: - name: sentiment_score_min reason: must be less than or equal to sentiment_score_max FeedbackRecordsUnauthorized: description: >- Unauthorized — no credentials, or an API key that does not exist or is revoked. Answered by the gateway as `text/plain`, before the request reaches the feedback store. content: text/plain: schema: type: string example: Unauthorized FeedbackRecordsForbidden: description: >- Forbidden. Answered by the gateway as `text/plain` whenever the request cannot be authorized: the dataset does not exist, is archived, or is not assigned to a workspace the API key has the required permission on; the organization lacks the Unify Feedback entitlement; a mutation was attempted with an API key on a dataset shared by more than one workspace; or the record id does not exist. Existence is not leaked: an unknown dataset or record is refused here rather than with a 404. The single-record operations are the one exception, and only for a minute after a delete — see their `404`. content: text/plain: schema: type: string example: Forbidden FeedbackRecordsPayloadTooLarge: description: >- Payload Too Large. The gateway buffers the request body in order to authorize it, and the proxy in front of it stops reading at 1 MiB — Envoy's `maxRequestBytes` on Formbricks Cloud, Traefik's `maxbodysize` in the Docker install. The refusal is therefore generated by that proxy rather than by the application, so neither its media type nor its body is part of this contract; branch on the status alone. FeedbackRecordsInternalServerError: description: >- Internal Server Error from the feedback store (`application/problem+json`, `code: internal_server_error`). content: application/problem+json: schema: $ref: '#/components/schemas/HubProblem' FeedbackRecordsServiceUnavailable: description: >- Service Unavailable. From the **gateway** (`text/plain`): the record's dataset could not be resolved because the feedback store did not answer, the authorization service itself was unreachable, or the record's stored `tenant_id` is not a dataset id this API can address. From the **feedback store** (`application/problem+json`, `code: service_unavailable`): the operation needs a capability that is not configured on this instance — semantic search and similar records require an embedding model. content: text/plain: schema: type: string example: Feedback record lookup failed application/problem+json: schema: $ref: '#/components/schemas/HubProblem' securitySchemes: apiKeyAuth: type: apiKey in: header name: x-api-key description: > Management API key; must include **workspaceId** as an allowed workspace with read, write, or manage permission. ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.