Home/Docs/Vapi

Capture from Vapi

Vapi records caller and assistant on separate channels already. Give capture the call id and your API key: it pulls that stereo recording and scores it offline, or hand it a WAV you already downloaded.

The recording you capture feeds a five-dimension conversation-test: separated caller and agent channels are what the Conversation (turn-taking) and Speech (timing) dimensions read, and the same file drops into hotato test run beside a transcript and trace for the Outcome and Policy checks.

Stack	Automatic pull	Separate channels	Contract-grade default	Network use
Vapi	Yes: `capture` fetches by call id	Yes, out of the box	Yes	One call: fetch the recording
Twilio	Yes: `capture` fetches by recording SID	No by default: must request `RecordingChannels=dual` at creation	No, opt-in	One call: fetch the recording
Retell	Yes: `capture` fetches by call id	Yes, once call recording is turned on	Opt-in: enable recording first	One call: fetch the recording
LiveKit	No: egress runs in your own infra	Yes, if egress records both tracks	Opt-in: configure egress for two tracks	None from Hotato
Pipecat	No: the `setup` recorder runs in your pipeline	Yes, the scaffold writes caller ch0 / agent ch1	Yes, once the recorder is added	None: local recorder, no network call

Automatic pull: does hotato capture fetch it for you. Separate channels: is caller/agent separation what you get. Contract-grade default: is that separation on without extra setup. Network use: what capture calls out to, not your infrastructure.

Enable dual-channel recording

Vapi’s recorded calls are stereo by default: once call recording is on for the assistant, artifact.recording.stereoUrl is a two-channel WAV already, customer on channel 0, assistant on channel 1. No separate dual-channel switch to find, unlike Twilio or LiveKit.
Recording off for this assistant? Turn it on; the stereo file is what you already get, no further config.

Same overlap, two recordings. Two-channel keeps caller and agent on separate tracks, so the overlap has an owner. Mono-mixed sums both onto one track: the overlap is still there, but nothing says who caused it. Turn on dual-channel recording upstream, above.

Score it: auto-pull

Set VAPI_API_KEY + call ID: capture pulls it (only network call: that download), scores offline. --stack vapi = fix knobs; --expect yield/hold = interruption vs. backchannel.

bash

export VAPI_API_KEY=...
uvx hotato capture --stack vapi --call-id <id>

# no key? prove the capture→score loop on a bundled call
uvx hotato capture --stack vapi --demo

Or score a WAV you already have

Downloaded it? Confirm two-channel (e.g. ffprobe): caller on channel 0, assistant on channel 1.

bash

uvx hotato run --stereo call.wav --stack vapi --expect yield

Single mixed track? --stereo and --caller/--agent need real per-party audio; a mono file there is a hard error. The opt-in [diarize] extra scores mono anyway by separating it first: run --mono call.wav --diarize runs a local diarizer, then gates on confidence. High separability scores as diarized-mono; low is stamped indicative only; a non-separable file still exits 2. Never equivalent to a true dual-channel recording. --allow-mono lives on capture, but doesn’t reach Vapi: its capture path only pulls the verified stereo artifact. Turn on dual-channel recording upstream (above) instead.

Pulling more than one call?

hotato connect vapi --api-key <key> stores the key once, locally (0600). hotato pull --stack vapi --since 7d bulk-fetches recent recordings via Vapi’s GET /call list endpoint; hotato sweep pulls, then ranks the worst turn-taking moments across all of them in one offline dashboard. Same stereo artifact, same offline scoring, just more calls at once. Every stack’s pull support →

The dual-channel gate

Separated caller and agent audio is what makes overlap attributable; speech activity comes from exposed VAD thresholds, frame-inspectable. The scorer takes one two-channel WAV or two aligned mono WAVs.