other · Listed · Found · 9 endpoints · Not used for Gateway
FishAudio OpenAPI
FishAudio OpenAPI
MPP endpoints published at https://api.fish.audio.
Not used for Gateway
Listed — live manifest, not yet verified · not used for Gateway Still catalogued — dead/unreachable knowledge is index value. (Listed — live manifest, not yet verified · not used for Gateway)
Indexed from this operator's public /.well-known/x402.json. Found is not operator-owned and is not attested. Claim or opt out.
Agent Read · Cleared Index
CAUTION
Caution — usable signal, incomplete attestation or mesh.
confidence
67%
source
signal
Index before you pay. Same payload for agents:
GET /api/cleared/agent-read?slug=fishaudio-openapi-hzxg
When to call
- MPP endpoints published at https://api.fish.audio.
Risks
- Found — not operator-owned; claim status unknown.
- No Cleared settlement receipt on file yet.
- Endpoint response check pending or failed.
- No Gateway traffic yet — market share unproven.
Price posture
9 endpoints — confirm price on manifest before pay.
Category · Gateway
other · no Gateway routes yet — early / unproven on Cleared market share.
Endpoint hints
POST /v1/agent/sessionsStart a conversation session with an agent and receive a join token for the session transport (currently LiveKit WebRTC). Authenticate with an API key to start
POST /v1/agent/agents/{agent_id}/tests/runStart a batch that runs the agent's attached tests against its current draft, one run per test and repeat. A simulation test repeats `simulation.repeat_count` t
POST /v1/agent/phone-numbers`provider` discriminates two variants. `twilio` buys a number from the managed inventory: it lands in your default workspace, any `agent_id` you bind must live
POST /v1/agent/phone-callsPlace an outbound call from one of your phone numbers. Platform numbers dial US, Canada or Japan destinations, and a domestic trunk 0 after +81 (e.g. +81080...)
POST /v1/asrPOST /v1/asr at $Metered.
POST /v1/ttsPOST /v1/tts at $Metered.
POST /v1/tts/stream/with-timestampPOST /v1/tts/stream/with-timestamp at $Metered.
GET /v1/tts/live/with-timestampStreaming text-to-speech over a WebSocket that returns word-level timestamps alongside the audio. Connect to `wss://api.fish.audio/v1/tts/live/with-timestamp`;
Evidence (Cleared)
- → Intake listed · not used for Gateway
- → Trust 55/100 · fail · tier listed
- → Protocol mpp
- → Manifest reachable · schema valid
- → Found listing — indexed from public x402.json, not operator-attested.
Endpoints
Create Agent Session
$MeteredPOST https://api.fish.audio/v1/agent/sessionsStart a conversation session with an agent and receive a join token for the session transport (currently LiveKit WebRTC). Authenticate with an API key to start sessions with any agent in your team; without credentials only agents published as public are reachable, and the request `Origin` must match the agent's allowed origins. `language` (and `overrides.language`) accepts `en`, `ja`, `zh`, `ko`, `es`, `fr`, `de`, `pt`, `it`, `nl`; anything else is 422. Omit it to use the agent's configured speaking language. `overrides.asr.model` picks the speech recognition model for this session, nested like the agent config's `asr` section and with the same values as `asr.model`. It needs an API key. `overrides.llm.fallback` replaces the agent's `llm.fallback` for this session, with the same shape and rules; each `custom` model needs its `api_key`. It needs an API key and `llm.fallback` in the agent's `overrides_allowed`.
Run Tests
$MeteredPOST https://api.fish.audio/v1/agent/agents/{agent_id}/tests/runStart a batch that runs the agent's attached tests against its current draft, one run per test and repeat. A simulation test repeats `simulation.repeat_count` times, and a `repeat_count` in the body repeats every selected test that many times instead. A batch starts at most 1000 runs. Runs are queued and finish in the background, simulations take minutes. Poll `GET /v1/agent/agents/{agent_id}/test-batches/{batch_id}` until `completed` is true, then gate on `pass_rate`, which leaves out `error` runs. Each simulation run that ends `passed` or `failed` is billed: a platform fee per agent reply plus the LLM tokens of the agent, the simulated user and the judge. `error` runs are not billed. A batch with a simulation test returns 402 when the account has no trial balance or API credit left, and 403 when the team's agreement does not allow agent usage.
Purchase or Import Phone Number
$MeteredPOST https://api.fish.audio/v1/agent/phone-numbers`provider` discriminates two variants. `twilio` buys a number from the managed inventory: it lands in your default workspace, any `agent_id` you bind must live there too, and billing is the monthly price charged in daily slices; the first day is charged before anything is bought (402 costs you nothing) and the daily run advances it from there. `sip` imports a number you already own at your carrier: point your trunk's origination at our SIP host, provide inbound authentication (digest and/or source CIDRs) and optionally a termination host so the number can place calls; nothing is rented and there is no monthly fee. Either way, 409 means the number is already on the platform; 502 means the provider or trunk provisioning refused, and the number stays visible with status `error` and is safe to release.
Create Phone Call
$MeteredPOST https://api.fish.audio/v1/agent/phone-callsPlace an outbound call from one of your phone numbers. Platform numbers dial US, Canada or Japan destinations, and a domestic trunk 0 after +81 (e.g. +81080...) is accepted and normalized to E.164 (+8180...). Imported SIP numbers dial through their own termination and may attach custom INVITE headers with `sip_headers`. Returns immediately with the session queued for dialing. Subscribe to the `phone_call.dial_finished` webhook or poll `GET /v1/agent/sessions/{session_id}` for the dial outcome and the leg's `sip_call_id`. Ringing is never billed: metering starts when the callee answers. Errors carry a machine-readable `reason` (e.g. `destination_not_allowed`, `insufficient_credit`, `concurrency_limit_exceeded`).
Speech to Text
$MeteredPOST https://api.fish.audio/v1/asrText to Speech
$MeteredPOST https://api.fish.audio/v1/ttsText to Speech Stream with Timestamps
$MeteredPOST https://api.fish.audio/v1/tts/stream/with-timestampText to Speech Live with Timestamps (WebSocket)
$MeteredGET https://api.fish.audio/v1/tts/live/with-timestampStreaming text-to-speech over a WebSocket that returns word-level timestamps alongside the audio. Connect to `wss://api.fish.audio/v1/tts/live/with-timestamp`; this is a WebSocket upgrade, not a plain HTTP request. Use it when you need to feed text in as it is produced — an LLM token stream, a live caption feed — and want word timings for lip-sync, subtitles or highlighting. For a one-shot request use `POST /v1/tts/stream/with-timestamp` instead. ## Handshake Send the same headers as any other API call: - `Authorization: Bearer <your-api-key>` - `model: <model>` — optional, defaults to `s2.1-pro`. ## Framing Every frame in both directions is a MessagePack-encoded **binary** frame holding a map with an `event` key. Text frames are ignored by the server and should be ignored by your client. ## Client to server The inbound protocol is identical to `/v1/tts/live`: - `start` — opens the session. Send it once per session, before any other event. Carries a `request` object with the same fields as the `POST /v1/tts` body; leave `text` empty and stream the words with `text` events. After a `finish` you may send another `start` to run a new session on the same socket. - `text` — appends a text fragment to the session buffer. - `flush` — forces buffered text through generation without waiting for the chunk to fill. The session stays open. - `stop` — ends the session. The server drains the buffer and answers with a final `finish` event. ## Server to client - `audio` — one audio chunk. Beyond the `audio` bytes it carries `content`, `alignment`, `chunk_seq` and `chunk_audio_offset_sec`; these last three are what this endpoint adds over `/v1/tts/live`. Concatenate every `audio` payload in arrival order to reconstruct the full audio. Note that `audio` here is raw MessagePack `bin`, not the base64 string used by the SSE endpoint. - `finish` — terminal event, with `reason` of `stop` or `error`. - `error` — a request-level failure such as an invalid `start` request or a concurrency refusal; the socket closes afterwards. ## Reading the alignment `alignment` is a **cumulative snapshot** for its `chunk_seq`, not a delta: when it is non-null, replace the alignment you stored for that chunk rather than appending to it. A `null` alignment means nothing new has been committed for the chunk yet, which is normal early on. Segment times inside `alignment` are relative to the start of their chunk. `chunk_audio_offset_sec` is the chunk's own start time within the full audio, so the absolute time of a word is `chunk_audio_offset_sec + segment.start`. Both `chunk_seq` and `chunk_audio_offset_sec` are monotonically non-decreasing. A frame may be labelled with the previous chunk's `chunk_seq` while carrying the next chunk's audio bytes, so never use `chunk_seq` to attribute audio bytes — only alignment.
Voice Design
$MeteredPOST https://api.fish.audio/v1/voice-designChecks
reachable
valid
2026-11-19T16:48:32.994Z
Intake listed · Listed — live manifest, not yet verified · not used for Gateway · probed 0/3 · No settlement evidence found in chain signals.
Claim to manage this card — or go straight to Cleared attestation ($49/mo).
Claim listing