7.4 billion tokens per month. 34 free LLM providers. 635 free model endpoints. All behind one /v1 endpoint, plus any custom OpenAI-compatible endpoint. Smart routing, automatic failover, encrypted keys. Personal experimentation only.
Free.ai
The Free.ai integration currently supports Kokoro text-to-speech only through
POST /v1/audio/speech. Model rows are distributed by the signed catalog; there
are no bundled seeds or changes to premium/free catalog timing.
Get a key at https://free.ai/account/#api-keys. The confirmed free account tier has 30,000 shared internal tokens per day and 1,000 API requests per month, without a card. These are account allowances, not a grant per model. Premium external models require purchased tokens and are not included by this adapter.
Use model kokoro, voice af_heart and response_format: "wav". If the format is
omitted, WAV is returned. Standard OpenAI voice names map to af_heart; native
voice names pass through. Other audio formats and untested models are rejected
before generation. The adapter calls Free.ai's native /v1/tts/, then retrieves
the WAV from its fixed output host without forwarding the credential or following
redirects. It checks the output bytes before returning them.
Free.ai can queue requests under load. This adapter reports those as a temporary
502 (upstream_queued), not successful audio. It does not follow
the browser-session queue URL or resubmit the generation automatically. Check
the provider's account history before retrying a queued request.
Credential health
Live controls on 2026-10-11 found that the model roster, invalid-payload errors,
invalid-model errors and even a successful Kokoro synthesis returned the same
status for valid and bogus credentials. Those endpoints cannot prove that a key
is valid. Automatic health checks therefore skip credential validation entirely without
spending quota or recording a misleading validation error. Existing unknown
and healthy key states remain usable for inference; their status is not proof
that the credential is valid. The supplied key is still sent on actual
generation requests; this integration does not advertise anonymous access.
Kokoro generated valid WAV files during live testing. Two tested Qwen aliases returned the same backend identity under conflicting advertised names; the other chat, image, translation and transcription candidates failed or timed out. None of those routes is included here.
References: https://free.ai/api/ and https://free.ai/developer/. Some older pricing text conflicts with these pages; the logged-in free account explicitly confirmed the shared daily pool and monthly API cap, with zero purchased tokens.