PyAI

Partial degradation

Last checked 2026-08-07T09:50:16.915Z · auto-refreshes every 30s

Products & endpoints

Platform

Edge, auth, catalog, console, and website.

Operational
  • API edge Gateway auth + routing (api.pyai.com). GET /healthz OperationalHTTP 200 · 46ms TTFB
  • API contract The live OpenAPI spec served at the edge. GET /openapi.json OperationalHTTP 200 · 61ms TTFB
  • Agent index The llms.txt index for AI coding agents. GET /llms.txt OperationalHTTP 200 · 55ms TTFB
  • Models catalog Model list (any active key). GET /v1/models OperationalHTTP 401 · 46ms TTFB
  • Voices catalog Voice library (any active key). GET /v1/voices OperationalHTTP 401 · 45ms TTFB
  • Console Developer console + key management. OperationalHTTP 200 · 94ms TTFB
  • Website pyai.com. OperationalHTTP 200 · 92ms TTFB

Hear, speech-to-text

Batch + streaming transcription, diarization, and Cue grounding. pyai-hear

Operational
  • Batch transcription Whisper-compatible speech-to-text (live transcription canary). POST /v1/audio/transcriptionsscope: hear:transcribe OperationalHTTP 200 · 305ms TTFB
  • Streaming transcription Low-latency partials; Cue with grounding (engine deep health). GET /v1/audio/transcriptions/stream (WS)scope: hear:stream OperationalHTTP 200 · 42ms TTFB
  • Async jobs Batch STT job queue (engine deep health). POST/GET /v1/transcription/jobsscope: transcribe:jobs OperationalHTTP 200 · 86ms TTFB

Speak, text-to-speech

Synthesis, voice cloning, and prompt-to-voice design. pyai-voice

Degraded
  • Speech synthesis Text-to-speech, streaming lane (live synthesis canary; latency shown is time-to-first-audio). POST /v1/audio/speechscope: voice:synthesize OperationalHTTP 200 · 183ms TTFB
  • Voice cloning Clone synthesis (live canary on a pinned voice). POST /v1/voice/clones/gaurav_test/synthesizescope: voice:clone DegradedHTTP 200 · 7752ms TTFB
  • Voice design Prompt-to-voice synthetic design. POST /v1/voice/designscope: voice:design OperationalHTTP 401 · 40ms TTFB

Omni, agentic voice

Realtime speech-to-speech duplex (native + OpenAI-compat). pyai-omni-realtime

Operational
  • Omni realtime (native) Native duplex speech-to-speech (engine health). GET /v1/omni (WS)scope: omni:session OperationalHTTP 200 · 50ms TTFB
  • Omni realtime (OpenAI-compat) OpenAI Realtime-compatible surface (realtime-shim health). GET /v1/realtime?model=pyai-omni-realtime (WS)scope: omni:session OperationalHTTP 200 · 45ms TTFB

AMD, answering-machine detection

Realtime who-or-what-answered detection for outbound calls. pyai-amd

Operational
  • Realtime detection Twilio Media Streams drop-in (live handshake probe; keyless by design). GET /v1/amd/stream (WS)scope: amd:detect OperationalHTTP 101 · 41ms TTFB
  • Config + call records Decision history and the aggressiveness dial (edge + control plane). GET /v1/amd/calls, POST/GET /v1/amd/configscope: amd:read OperationalHTTP 401 · 46ms TTFB
  • End-to-end decision canary Real key, real decision, durable record — the continuous functional canary's verdict. POST /v1/amd/stream (WS) + GET /v1/amd/calls/{id}scope: amd:detect OperationalHTTP 200 · 150ms

Multilingual

Per-language speech model quality across supported languages.

Degraded
  • Per-language model check Hourly check that every supported language is served by its intended speech model. DegradedHTTP 200 · 170ms

Recent history

Incident · major

AMD realtime detection unavailable

  • resolved AMD realtime detection is fully recovered. New connections to /v1/amd/stream are accepted and authenticated normally, and verified end to end. The status page now probes the AMD connection handshake directly, so a recurrence shows as down here within seconds. 2026-08-06T15:10:00Z
  • identified All connections to AMD realtime detection (wss://api.pyai.com/v1/amd/stream) have been failing since 2026-08-04 07:12 UTC. No call records or decisions were produced during the window; configuration and call-history reads were unaffected. Cause was an internal credential mismatch introduced during a routine engine update. 2026-08-06T13:47:00Z
Incident · major

Speak TTS synthesis unavailable

  • resolved Speak synthesis and clone synthesize recovered. Platform status page now uses live POST canaries (not just /healthz/*) so future synthesis outages surface as down. Monitoring. 2026-06-25T20:00:00Z
  • investigating POST /v1/audio/speech returns 503 (TTS synthesis unavailable) and clone synthesize returns 500. Speak and Clone share the pyai-voice-v3 engine. Gateway deep-health routes still report operational, Platform deployed live synthesis canaries on the status page and escalated to the engine team. 2026-06-25T18:00:00Z

Subscribe to updates

Get incident & maintenance notifications by email, or plug the feed into your tools, no account needed.

Or subscribe via feed:

RSS / Atom feed Incidents JSON Status JSON History JSON