> ## Documentation Index > Fetch the complete documentation index at: https://docs.textsetu.com/llms.txt > Use this file to discover all available pages before exploring further. # Translate with AI (write) > Machine-translate the project's candidate keys into one or more target languages with the resolved engine. Runs asynchronously — poll the returned `jobId`. Candidates are filtered in SQL before any model sees a string (`onlyEmpty` / `onlyOutdated` / tag filters / `keyPrefix` / explicit `keyIds`). With `?branch=` the run targets a branch instead of main. Writes follow the project's approval workflow exactly as in the app. **Constraints:** - Metered against the org's AI credits when running on a TextSetu platform key; free on a BYOK key. - `autoApprove` is honored only for callers that can approve; otherwise writes land as proposals. ## OpenAPI ````yaml /openapi/textsetu.json post /projects/{projectId}/ai/translations openapi: 3.1.0 info: title: TextSetu API version: 1.2.0 description: >- Public REST API for TextSetu. Authenticate with a Personal Access Token (tsu_pat_…) or a project token (tsu_proj_…) via the `Authorization: Bearer ` header. **Authorization is permission-based.** Every token carries an explicit allow-list of RBAC permission keys (e.g. `translation_read`), and each endpoint declares the permission it requires (`x-permission`). A token's effective access is its underlying authority ∩ its allow-list — a PAT is further bounded by its owner's role-based permissions, so it can only narrow, never exceed, what the owner already has. A project token is bound to a single project. There are no coarse read/write/manage scopes. **Response envelope.** Every endpoint returns `{ "success": true, "data": … }` on success and `{ "success": false, "error": { "code", "message" } }` on failure. **Rate limit.** 600 requests/minute, keyed by token (or by IP when unauthenticated). Exceeding it returns 429 with the standard error envelope. servers: - url: https://api.textsetu.com/api/v1 description: Production - url: /api/v1 description: This server (self-hosted / same-origin) security: - bearerAuth: [] tags: - name: Projects description: >- Read project metadata — settings, key separator, and the approval/branching flags that shape how the other endpoints behave. - name: Languages description: >- List and manage a project's target languages. Each language is referenced by its BCP-47 code; the source language is flagged separately. - name: Translations description: >- Manage a project's translations: create, read, update, and delete translation keys — the identifiers your app looks up, carrying metadata (description, tags, screenshot) — and set the translated value for a key in a given language. Value writes honor the project's approval workflow — non-approver writes land as `pending_review`. - name: Stats description: >- Per-language completeness and approval progress for a project — the numbers behind the dashboard. - name: Import/Export description: >- Bulk-load source files into a project or export the current translations. Supports the same file formats as the web app. - name: Branches description: >- Work on translations in isolation and merge them back. Branch reads/writes mirror the main endpoints but stage changes as a diff until merge. See the [branching guide](/docs/guides/branches). - name: Glossary description: >- Manage term bases — approved terminology and its translations — and check a string against them. See the [glossary guide](/docs/guides/glossary). - name: Translation Memory description: >- Reusable translation stores. Fuzzy-match and concordance-search prior translations to reuse them. See the [translation memory guide](/docs/guides/translation-memory). - name: AI description: >- Machine-translate a project with the org's configured AI engine (brand voice, glossary/TM grounding, and an optional review pass). Runs asynchronously — start a run and poll the returned `jobId`. - name: Webhooks description: >- Get notified when a project's translations change, instead of polling. Register an HTTPS endpoint, subscribe it to the events you care about, and TextSetu POSTs a signed payload whenever one occurs. Payloads follow the [Standard Webhooks](https://www.standardwebhooks.com) spec — verify the `webhook-signature` header with any compatible library. Delivery is at-least-once and unordered, so treat `webhook-id` as an idempotency key. Failed deliveries are retried five times over roughly seven hours, and every attempt is inspectable and replayable via the delivery log. - name: Meta description: Service-level and cross-cutting endpoints. paths: /projects/{projectId}/ai/translations: post: tags: - AI summary: Translate with AI (write) description: >- Machine-translate the project's candidate keys into one or more target languages with the resolved engine. Runs asynchronously — poll the returned `jobId`. Candidates are filtered in SQL before any model sees a string (`onlyEmpty` / `onlyOutdated` / tag filters / `keyPrefix` / explicit `keyIds`). With `?branch=` the run targets a branch instead of main. Writes follow the project's approval workflow exactly as in the app. **Constraints:** - Metered against the org's AI credits when running on a TextSetu platform key; free on a BYOK key. - `autoApprove` is honored only for callers that can approve; otherwise writes land as proposals. parameters: - name: projectId in: path required: true schema: type: string description: Project id (UUID) or id-embedded path slug. - name: branch in: query schema: type: string description: >- Target an open translation branch instead of main. Required when the project's main branch is protected. requestBody: required: true content: application/json: schema: type: object properties: targetLanguageCodes: minItems: 1 type: array items: type: string minLength: 2 description: Target BCP-47 language codes to translate into. examples: - - fr - de keyIds: description: >- Restrict to these key ids. Omit to consider every key in the project. type: array items: type: string onlyEmpty: description: >- Translate only keys with no existing value in the target language. type: boolean onlyOutdated: description: >- Include values flagged outdated (the source changed after the value was last set). type: boolean includeTags: description: Only keys carrying at least one of these tags. type: array items: type: string excludeTags: description: Skip keys carrying any of these tags. type: array items: type: string keyPrefix: description: Only keys whose id starts with this prefix, e.g. `home.`. type: string autoApprove: description: >- Approvers only: apply directly instead of proposing when the project requires approval. Ignored for callers without approval rights (the write still lands as a proposal). type: boolean review: description: >- Run the engine's review pass on this run. Omit to follow the engine default. type: boolean batch: description: >- Use the provider's asynchronous batch path (~50% cheaper, higher latency). type: boolean required: - targetLanguageCodes responses: '202': description: The queued job's id and initial status. content: application/json: schema: type: object required: - success - data properties: success: type: boolean const: true data: type: object properties: jobId: type: string description: >- Id of the queued AI translation job. Poll the status endpoint with it. examples: - job_a1b2c3d4 status: type: string description: Initial job status (usually `queued`). examples: - queued required: - jobId - status additionalProperties: false '400': description: >- AI is not available for this project, no model is configured for a pair, or the target/branch is invalid. '401': description: The token has no associated user to attribute writes to. '402': description: >- Insufficient AI credits to cover the run (platform-key runs only; BYOK is never blocked). '403': description: Forbidden / main branch protected. components: securitySchemes: bearerAuth: type: http scheme: bearer description: >- API token: tsu_pat_… (PAT) or tsu_proj_… (project token). The token's granted RBAC permissions determine access; see each operation's `x-permission`. ````