Deepl

REST API at api-free.deepl.com (free) or api.deepl.com (pro). Text translation: POST to /v2/translate with text, target_lang, and optional source_lang, formality, context, and glossary_id. Supports multiple text segments in a single request (array of strings). Response returns translations with detected source language. Document translation is async: POST document to /v2/document, poll status, retrieve translated document — preserving original formatting. Glossary management: create glossaries mapping source→target terms for consistent translation. Usage endpoint reports character consumption. The API is clean and purpose-built — no unnecessary complexity. XML and HTML tag handling preserves markup in translations. Split sentence control enables fine-grained segmentation.

API errors return JSON with message field. Standard HTTP status codes: 403 for auth errors, 456 for quota exceeded, 429 for rate limits. The 456 status code is unique to DeepL (quota exhausted) — agents must handle this specifically. Rate limits are generous but undocumented in exact numbers — the API returns 429 with Retry-After. Document translation polling may return 'translating' status for large documents — agents should poll with reasonable intervals. Translation quality is generally high but can struggle with highly technical or domain-specific content — glossaries mitigate this. The API is stateless for text translation (no async polling needed), which simplifies error handling. Character counting includes source text length, not translated text.

DeepL provides machine translation with quality that consistently outperforms Google Translate and Amazon Translate for European language pairs. For agents, the API is straightforward: send text, receive translation. The key capabilities are text translation (up to 128KB per request), document translation (PDF, DOCX, PPTX, XLSX, HTML, TXT — preserving formatting), glossary management (enforce consistent terminology), and formal/informal tone control. The free tier offers 500,000 characters/month — generous for development and moderate production use. Language detection is automatic. The context parameter enables agents to provide surrounding context for improved translation accuracy. For agents localizing content, translating user inputs, or building multilingual features, DeepL's quality advantage is significant. The limitation: Asian language support is growing but European pairs remain the strength.

API key authentication via DeepL-Auth-Key header. Keys are generated per account. Free tier uses a different API hostname (api-free.deepl.com) than Pro (api.deepl.com). No fine-grained key scoping — the key grants full API access. No OAuth. Usage is tracked per account with monthly character limits. Free tier: 500K characters/month. Pro tier: pay-per-character beyond base allocation. For agents, the free tier is enough for development and testing. The separate hostnames for free vs. pro means agents must configure the correct base URL. No IP restrictions. No temporary credentials. The auth model is simple — appropriate for a focused translation API.

Documentation at developers.deepl.com/docs is clean and well-organized. API reference covers all endpoints with parameters, response examples, and error codes. The documentation is concise — the API surface is small enough that the entire reference can be read in 30 minutes. Official client libraries for Python, Node.js, .NET, and Java handle auth, error handling, and document translation polling automatically. The libraries are well-maintained by DeepL. The glossary documentation explains usage patterns clearly. The free tier documentation is explicit about limitations. Community resources are limited compared to Google Translate but the official docs are sufficient. For agents, the combination of clean API design and good client libraries makes integration fast.