From a call to a breakdown
How CLUE turns a call into a finished breakdown
You set the phone system up once. After that every finished conversation runs automatically: recording, transcript, analysis, dashboard. Every step below comes with real timings.
5 calls a month. $0. No card.
Four steps
What you do, and what the system does
Email or Google, then 5 short steps: role, team size, main goal and current stack. The starting dashboard is assembled around your role.
about 3 minutesSettings, the integrations tab, your provider, a key and a webhook. Then you link each rep's extension so a call lands in the right profile.
about 15 minutesThe phone system sends the recording, CLUE fetches it, encrypts it, turns it into text with the rep and the customer split apart, and runs it through 10 sections of analysis.
1 to 5 minutes after the callThe rep gets their breakdown first. The lead sees the team overview, the weak stages and the top objections. The one-to-one starts at a timecode, not at an impression.
the same dayStep 2, in detail
Connecting the phone system, step by step
This is the only technical step in the whole rollout, and it happens in the interface, without a developer.
This section is available to the workspace admin
Binotel, UniTalk, Ringostat and others
Encrypted with AES-256-GCM the moment it is saved
One line in your phone system's settings
So a call lands in the right profile
Step 3, in detail
7 stages of call processing
Scroll to follow a call from the webhook to a finished breakdown. If a stage fails, the call returns to the queue rather than disappearing.
- PENDINGCall received
- DOWNLOADING_AUDIORecording downloading
- UPLOADING_TO_S3Encrypted storage
- TRANSCRIBINGTranscription and speakers
- ANALYZING10 sections of analysis
- INDEXINGIndexed for search
- COMPLETEDDashboard updated
Your phone system tells CLUE the moment a call ends
The webhook arrives on its own, the record is created, and the call joins the queue call-processing.
CLUE fetches the recording from the provider
The audio is pulled from the provider's link. A slow provider does not break processing: the queue waits and retries.
The audio goes into encrypted storage
Backblaze B2, S3 compatible. Encrypted at rest and in transit, with companies' data isolated.
Speech becomes text, with the speakers separated
Three transcription tiers with fallback: Whisper large-v3 on GPU, Deepgram, and local Whisper. If the first tier is unavailable, the next one takes over.
The model breaks the conversation into 10 sections
Quantitative measures, qualitative judgements, structure, objections, recommendations, strengths, areas to work on, the scoring formula, stages, and the summary.
- Objection handled
- Next step captured
- Risk: a competitor came up
The conversation becomes searchable
The transcript, the keywords and the analysis JSON go into Postgres indexes. Search works across the whole call history.
The breakdown is in the dashboard
The lead sees the team overview, the rep sees their own conversation. Notifications can go to Slack or Telegram.
How long it takes
Real processing times
Get started
15 minutes to connect — after that the system runs itself
Free: 5 calls a month, no card.