For AI agents

Entry points

• /llms.txt

  A map of every documentation page with links to their Markdown versions. Start here.

  https://lumean.app/llms.txtCopy the link
• llm.md

  The whole API in one file, with sections and end-to-end recipes. Use it when you need full context. The file is in Russian only.

  https://api.lumean.app/docs/llm.mdCopy the llm.md link
• .md

  The Markdown version of any page: replace /developers/docs/ in the URL with /developers/docs/md/. Use it when you need one topic.

  https://lumean.app/en/developers/docs/md/<path>
• OpenAPI

  The REST contract: operations, parameters and schemas. Use it to generate a client and exact types. Available in Russian and English.

  https://api.lumean.app/docs/public-openapi.en.jsonCopy the link
• AsyncAPI

  WebSocket channels and events. Use it to subscribe to order statuses instead of polling.

  https://api.lumean.app/docs/public-asyncapi.en.jsonCopy the link

Reading order

1. Read /llms.txt and pick the pages you need.
2. Fetch their Markdown versions, not HTML.
3. Take schemas and types from OpenAPI.
4. Send requests with the X-API-KEY header. The account owner creates the key in their account.

API rules

Agents get these rules wrong most often. /llms.txt and the Markdown version of this page carry the same rules in English.

• The voice is set in a template

  An order has no voice_id field. The voice is stored in the template’s config.tts_settings.voice_id, and the order refers to the template by template_id. To voice a single order with another voice, pass config_override.tts_settings.voice_id. The text to voice goes in input_text.

• Orders run asynchronously

  POST /orders returns 201 with status created and no result yet. Poll GET /orders/{order} or listen over WebSocket to follow the status. partially_completed is not a final status: some chunks failed, and you can retry them or cancel the order.

• result.files[] holds paths, not links

  POST /storage/url returns the download link, valid for one hour. Order files are kept for 7 days after the order finishes.

• Retrying an order creates a new order

  POST /orders/{order}/retry creates a new order with the same parameters and is billed again. Failed chunks are retried inside the original order: POST /orders/{order}/items/retry-failed. POST /orders has no idempotency key, so after a dropped connection, look for the order in the list first.

• A 429 has several causes

  It can be the request rate limit, the key’s token quota, a subscription limit, the concurrent task limit, or a pause after content policy blocks. The reason field names the cause. If the response has a Retry-After header, wait the time it gives. concurrency_limit_exceeded comes without it and lifts when your active orders finish. limit_exceeded is the subscription limit: retrying does not help until the period resets or the plan changes.

What models get wrong

The list from section 0 of llm.md. Add it to the agent’s context before the first request.