What it does
- Pulls call recordings from any phone system: HousecallPro, RingCentral, Dialpad, Twilio, OpenPhone, JustCall, your VOIP — pick one or several.
- Transcribes with speaker attribution so you can scan who said what at a glance.
- Generates a short AI summary tuned to your business (e.g. "customer agreed to Tuesday 10am, wants a generator quote").
- Pushes the transcript + summary back into your CRM as a job note or attachment, automatically.
- Catches missed action items ("send the address back", "text the quote", "schedule follow-up") so they don't fall through.
How it works
Whenever a call ends, the recording is fetched from your phone system. The audio runs through best-in-class speech-to-text (Whisper from OpenAI, or Deepgram for high-throughput), with speaker diarization so the transcript is tagged "CSR" vs "Customer" automatically.
An LLM (Claude or GPT-4) reads the transcript with your business context (job type, customer history) and writes a tight summary — including any commitments or follow-ups the call generated.
The full transcript and summary land in your CRM as a job note, an attachment, or a custom field — whichever your CRM supports. Usually within 30 seconds of the call ending.
The stack
OpenAI Whisper or Deepgram for transcription with speaker diarization. Claude (Anthropic) or GPT-4 (OpenAI) for summary. Python orchestrator. Provider-specific webhooks for fetching audio (HCP, RingCentral, Twilio, OpenPhone, etc.). Generic REST integration for pushing results back into any CRM.
Who it's for
Any team where calls are the heart of the business but call notes are the bottleneck. Service businesses, sales teams, customer success, intake operations — anyone who's ever asked "what did they actually say on that call?".