Logo

Complete developer guide and best practices to help you get started quickly.

Overall shape

Core concepts

An overview of how the API is shaped: two endpoint types (public lookup vs. key-protected business endpoints), the shared response structure, and the recurring limits you will encounter.

Two endpoint types

Path prefix indicates the auth type, so the URL alone tells you whether a key is required.

Public endpoints (no prefix)

No key required. Use these to look up supported languages, available models, and similar metadata before integrating. Safe to call from the browser — no sensitive data is exposed.

  • GET /translations/languages — supported languages for the translation module.
  • GET /translations/models — available translation models.
  • GET /audio/languages — supported languages for the speech recognition module.
  • GET /audio/models — available speech recognition models.

Key-protected endpoints (/api/v1)

Require X-API-Key in the HTTP header. These cover the core business operations: text translation, speech recognition, dictionary management, and so on.

  • POST /api/v1/translations/text — translate text, with optional streaming.
  • POST /api/v1/audio/transcriptions — speech recognition, accepting either Base64 or file upload.
  • /api/v1/dictionaries/* — dictionary management (vocabulary, forced replacement, batch operations).
  • GET /api/v1/usage — current-period account- and key-level usage percentages; consumes no quota.

How to get an API key

Create one on the API keys page in the developer console. Key creation and revocation are performed only through the console UI; no public management API is exposed.

Shared response shape

On success the result fields are returned directly (no data wrapper); translation and speech recognition responses also include usage (the current-period quota usage percentage, quota_percent.used). On failure it carries an error object with code, message, details, and request_id. Every error response includes a request_id, which lets us pinpoint the corresponding server log when you report an issue.

Recurring limits you will see

  • Text translation accepts up to 5000 characters per call; exceeding the limit returns 400 validation_error.
  • Speech recognition limits depend on the upload path: the JSON (Base64) path caps raw audio at 1 MB (413 audio_too_long beyond that); the multipart file upload path caps the file at 15 MB (413 file_too_large beyond that).
  • Dictionary batch operations accept up to 250 entries per call.

Keep reading

See “Error handling” for the full list of error codes, “Request ID and issue reporting” for the request_id workflow, and “SSE streaming” for the streamed event structure.