Score your Retell calls.
Give capture a Retell call id and your API key. Hotato pulls that call's multi-channel recording — the PII-scrubbed track first — and scores the turn-taking timing offline.
Auto-pull, by explicit call id.
GET https://api.retellai.com/v2/get-call/<id> over Bearer auth returns the per-party recording once the call ends. Hotato prefers scrubbed_recording_multi_channel_url, then recording_multi_channel_url.
Retell exposes no confirmed list-calls endpoint, so Hotato does not guess one: you pass the call id directly. There is no --since sweep for Retell — capture the calls you name. The plain mono recording_url is rejected unless you pass --allow-mono.
- connect0600 key
- →
- capture--call-id
- →
- multi-channelscrubbed
- →
- scoreno egress
- →
- scorecard5 dimensions
Hotato validates the downloaded WAV has two channels before it produces separated talk-over numbers. One party per channel, or it stops.
$ hotato connect retell$ hotato capture --stack retell --call-id <id>$ hotato capture --stack retell --call-id <id> --expect yieldNo key yet? hotato capture --stack retell --demo runs the loop on a bundled call. connect stores RETELL_API_KEY.
Two commands, one call at a time.
connect stores the key. capture pulls and scores one call by id. Because Retell has no list endpoint, you capture the calls you name rather than sweeping a window.
--expect yield treats a caller who starts talking as an interruption the agent should yield to; --expect hold treats it as a backchannel to talk through.
Score the calls you already ran.
Retell has no confirmed create-call API, so Hotato does not place Retell calls. It scores the recordings your live traffic already produced.
| Capability | Retell | Basis |
|---|---|---|
| Capture a recorded call | Yes | Auto-pull the multi-channel recording with your API key, explicit call id only |
| Drive a live call | Capture-only | No confirmed origination API — Hotato never places Retell calls |
Your key stays yours.
Local, 0600hotato connect retell stores RETELL_API_KEY once at file mode 0600 in ~/.hotato/connections.json. It is sent only to Retell's own API on a capture — never to Hotato.
ScopeHotato reads turn-taking timing — did_yield, seconds_to_yield, talk_over_sec — from speech energy. No speaker identification, diarization scoring, transcription grading, emotion or intent. Energy is not intent.