# ZYRAXON

> Free OpenAI-compatible AI provider. Ten ZYRAXON models behind one base URL,
> plus text-to-speech and speech-to-text. No API key to buy. Everything below is
> public and can be fetched without authentication.

## Start here

- [Provider overview](https://zyraxonai.lovable.app/providers): the whole API
  surface on one page — chat and streaming, the model catalogue, text to speech,
  speech to text, and integration steps for every host application. The page is a
  hub of *functions*, each of which has a page of its own.
- [Endpoints as JSON](https://zyraxonai.lovable.app/api/public/zyraxon/v1/endpoints):
  the same catalogue in machine-readable form, for an agent that would rather
  parse than render.
- [Model catalogue](https://zyraxonai.lovable.app/api/public/zyraxon/v1/models):
  the ten models, in the standard OpenAI list shape.
- [Integration catalogue](https://zyraxonai.lovable.app/api/public/zyraxon/v1/integrations):
  the exact values and snippets for OpenHands, Cursor, Cline, Aider, LiteLLM,
  LangChain, the Vercel AI SDK, Termux and others.

## One page per function

The provider is one base URL, but each thing it does has its own page. Fetch the
one you need rather than the whole overview.

- [Speech to text](https://zyraxonai.lovable.app/providers/speech-to-text) — if
  you are putting a microphone in your chat box, use this endpoint: you send
  real-time audio and it sends real-time text back.
- [Text to speech](https://zyraxonai.lovable.app/providers/text-to-speech) — send
  text, get audio back.
- [Images, audio, video and PDF](https://zyraxonai.lovable.app/providers/media)
  — every model takes every modality, inline or by URL.
- [Chat completions](https://zyraxonai.lovable.app/providers/chat),
  [The model line](https://zyraxonai.lovable.app/providers/models),
  [Tool calling](https://zyraxonai.lovable.app/providers/tools),
  [Capacity and errors](https://zyraxonai.lovable.app/providers/errors)
- Host applications:
  [Termux](https://zyraxonai.lovable.app/providers/termux),
  [Kali](https://zyraxonai.lovable.app/providers/kali-linux),
  [OpenHands](https://zyraxonai.lovable.app/providers/openhands),
  [Editors](https://zyraxonai.lovable.app/providers/editors),
  [Mobile apps](https://zyraxonai.lovable.app/providers/mobile),
  [SDKs](https://zyraxonai.lovable.app/providers/sdks),
  [Bots and automation](https://zyraxonai.lovable.app/providers/automation),
  [Voice in and out](https://zyraxonai.lovable.app/providers/voice)

## Endpoints

- Chat completions: `POST https://zyraxonai.lovable.app/api/public/zyraxon/v1/chat/completions`
  — OpenAI chat-completions semantics, with server-sent-event streaming.
- Models: `GET https://zyraxonai.lovable.app/api/public/zyraxon/v1/models`
- Provider status: `GET https://zyraxonai.lovable.app/api/public/zyraxon/v1/provider`
- Text to speech: `POST https://zyraxonai.lovable.app/api/public/tts` — 322 voices, 75 languages.
- Voices: `GET https://zyraxonai.lovable.app/api/public/tts/voices`
- Speech to text: `POST https://zyraxonai.lovable.app/api/public/stt` — one endpoint,
  two modes: send a recording for the whole transcript, or stream raw PCM and
  receive server-sent events as it is recognised. Accepts WAV, MP3, FLAC, OGG,
  WebM and raw PCM up to 100 MB; MP4/AAC is refused with a 400. `GET` on this URL
  returns its own contract, including the limits.
- Multimodal chat: the same `chat/completions` call takes an OpenAI content-parts
  array, so one message can mix text with an image, a recording, a video or a
  PDF — inline as base64 or as an `https://` URL the server fetches. A media turn
  is routed to a model that supports it, and the response carries
  `X-ZYRAXON-Input-Modalities`.

## Facts an agent should not guess at

- Authentication: none. Send any non-empty string as the API key if a client
  insists on one.
- The `language` field on the speech endpoint is required, in both modes. The
  recogniser does not detect language; it will answer confidently in the wrong
  one if asked, so the caller names the language rather than letting it guess.
- Model ids are ZYRAXON names (for example `zyraxon-apex`). They stay stable
  when the upstream pool behind them rotates.

## Optional

- [Documentation](https://zyraxonai.lovable.app/docs): longer guides, including
  speech, provider integrations and the endpoint reference.
- [Speech to text reference](https://zyraxonai.lovable.app/docs/provider-speech-to-text):
  every container, both modes, the 100 MB streaming path and the language rule.
- [Text to speech reference](https://zyraxonai.lovable.app/docs/provider-text-to-speech):
  the voice catalogue, sentence-by-sentence playback and the Markdown rules.
- [Multimodal chat reference](https://zyraxonai.lovable.app/docs/provider-multimodal):
  the content-parts array, routing by modality, and the rules for large media.