Skip to main content
Coldread

Core Concepts

Understanding how Coldread processes calls, analyzes them with individual AI calls, and builds contact profiles. Learn the fundamentals before diving into features.

Call Ingestion

Coldread connects to your VoIP system (Ringover, Aircall) via webhooks. When a call ends, your phone system sends the recording and metadata to Coldread automatically.

  • Webhooks trigger on call_ended events
  • Call audio and metadata are stored securely in R2
  • Processing begins within seconds of call completion
  • No manual uploads required

Transcription & Diarization

Every call is transcribed using ElevenLabs Scribe, which provides speaker diarization (identifying who said what). This forms the foundation for all downstream analysis.

  • Multi-speaker diarization (identifies up to 10 speakers)
  • Timestamp accuracy for each utterance
  • Supports UK/US/AU accents
  • Average processing time: 30-60 seconds per call

AI Analysis

After transcription, Coldread runs individual Gemini 2.5 Flash calls to detect the sales stage, classify the call type, evaluate custom tags, run compliance checks, and generate a call outcome summary.

  • Stage detection: automatically classifies where the call sits in your pipeline
  • Call type: categorizes the call (e.g. inbound inquiry, follow-up, demo)
  • Custom tags: each user-defined tag is evaluated individually per call
  • Compliance checks: each rule is checked individually against the transcript
  • Call outcome: a concise 15-word summary of the call result

Contacts & Profiles

Coldread automatically builds contact profiles from phone numbers and call history. Each contact has a timeline of all their calls with transcripts and analysis.

  • Automatic contact creation from phone metadata
  • Conversation timeline (all calls with that contact)
  • Call analysis linked to each conversation
  • Manual note-taking supported

Contact Intelligence

Contact Intelligence lets you ask natural language questions about any contact and get AI-powered answers based on their full conversation history.

  • Semantic search across all conversations
  • Natural language queries answered by AI
  • Results cite specific call timestamps

Privacy & Compliance

Coldread is built for GDPR compliance. Call recordings, transcripts, and Contact Intelligence data are all deletable, and data processing agreements (DPAs) are available.

  • Data deletion endpoint for GDPR compliance
  • Sub-processors documented in legal pages
  • No sensitive attribute inference (complies with AI provider ToS)
  • EU data residency available on request

How it all works together

1

Your VoIP system sends a webhook to Coldread when a call ends, including the recording URL and metadata (phone number, duration, direction).

2

Coldread downloads the recording, transcribes it with speaker diarization using ElevenLabs Scribe, and stores both the audio and transcript.

3

Individual Gemini 2.5 Flash calls analyze the transcript: stage detection, call type classification, custom tags, compliance checks, and a 15-word call outcome summary. Each feature is its own focused API call.

4

If the phone number doesn't exist, Coldread creates a new contact. If it does, the call is added to that contact's conversation timeline.

5

All analysis results are stored in the database and visible on your dashboard. Use Contact Intelligence to ask natural language questions about any contact's conversation history.

Dive deeper

Call Transcription

Speaker diarization, accuracy, and processing details

Compliance Monitoring

Custom compliance checks and GDPR features

Contact Intelligence

Natural language queries over conversation history

Team Management

Roles, permissions, and collaboration

Aircall Integration

Connect Aircall for automatic call ingestion

Ringover Integration

Connect Ringover for automatic call ingestion