Calls · Overview

Call intelligence — how it works

Ambrose pulls your sales calls from GHL, Retell, VAPI or JustCall, archives the recording, gets a transcript, scores the conversation against your scorecard, and writes one document per call that every Calls dashboard reads.

The pipeline

One call moves through four steps. Each is a skill your call AI agent can invoke:

Sweep — call-sweep-ghl, call-sweep-justcall, call-sweep-vapi or call-sweep-retell finds calls newer than the last run.
Intake — call-intake-anywhere archives the recording and gets a transcript.
Junk filter — call-junk-filter separates real conversations from voicemail, no-answers and wrong numbers.
Score — call-analyzer scores the conversation against the scorecard template that matches the call type, and tags the objections it heard.

Two rules worth knowing

The recording is always archived, even when the provider already gave us a transcript. The audio is the evidence behind a score — a rep disputing a coaching note needs to hear the moment — and a provider can expire its own copy at any time.

Transcription only runs when it has to. Retell, VAPI and JustCall usually return a transcript with the call. When they do, Ambrose uses it. AWS Transcribe only runs when there is no usable English transcript, because re-transcribing audio you were already handed costs money, is slower, and usually produces a worse transcript than the provider's.

The vault is the database

There is no second database. Everything lives in your agency's vault, and the dashboards read it directly:

call-intelligence/recordings/{callId}.wav
call-intelligence/recordings/{callId}.transcript.txt
call-intelligence/analyses/{callId}.json
call-intelligence/config/scorecard-templates.json
call-intelligence/config/objection-taxonomy.json

{callId} is the provider's own call id, unchanged — it is what joins the vault, the dashboards and the source system together.

An analysis that was not written to the vault did not happen. The Calls, Objections and Team pages read those documents and nothing else.

What you need before it works

A call AI agent under Calls → AI agents, set Live. A draft agent's routines do not run.
A connected call source — GHL, or JustCall/VAPI/Retell under Settings → MCP.
A routine on that agent, or a manual Run now.

Your scorecard templates and objection taxonomy are created for you the first time you open Call intelligence or create a call agent — you do not have to seed anything.

## Call intelligence — how it works

Ambrose pulls your sales calls from GHL, Retell, VAPI or JustCall, archives the recording, gets a transcript, scores the conversation against your scorecard, and writes one document per call that every Calls dashboard reads.

### The pipeline

One call moves through four steps. Each is a skill your call AI agent can invoke:

- Sweep — call-sweep-ghl, call-sweep-justcall, call-sweep-vapi or call-sweep-retell finds calls newer than the last run.

- Intake — call-intake-anywhere archives the recording and gets a transcript.

- Junk filter — call-junk-filter separates real conversations from voicemail, no-answers and wrong numbers.

- Score — call-analyzer scores the conversation against the scorecard template that matches the call type, and tags the objections it heard.

### Two rules worth knowing

**The recording is always archived**, even when the provider already gave us a transcript. The audio is the evidence behind a score — a rep disputing a coaching note needs to hear the moment — and a provider can expire its own copy at any time.

**Transcription only runs when it has to.** Retell, VAPI and JustCall usually return a transcript with the call. When they do, Ambrose uses it. AWS Transcribe only runs when there is no usable English transcript, because re-transcribing audio you were already handed costs money, is slower, and usually produces a worse transcript than the provider's.

### The vault is the database

There is no second database. Everything lives in your agency's vault, and the dashboards read it directly:

```
call-intelligence/recordings/{callId}.wav
call-intelligence/recordings/{callId}.transcript.txt
call-intelligence/analyses/{callId}.json
call-intelligence/config/scorecard-templates.json
call-intelligence/config/objection-taxonomy.json
```

`{callId}` is the provider's own call id, unchanged — it is what joins the vault, the dashboards and the source system together.

**An analysis that was not written to the vault did not happen.** The Calls, Objections and Team pages read those documents and nothing else.

### What you need before it works

- A call AI agent under Calls → AI agents, set Live. A draft agent's routines do not run.

- A connected call source — GHL, or JustCall/VAPI/Retell under Settings → MCP.

- A routine on that agent, or a manual Run now.

Your scorecard templates and objection taxonomy are created for you the first time you open Call intelligence (/call-intelligence) or create a call agent — you do not have to seed anything.