⚡ Voicemail triage — available now

Your AI Technician.
On Call 24/7.

ComFlow turns every missed call into a structured work item — and is becoming the AI co-worker that resolves them. Bring your own LLM, STT, and TTS. Benchmark everything. Own your stack.

View on GitHub See How It Works ↓
📞
Sarah Mitchell
Apex Manufacturing · +1 (555) 204-7823
Support High Urgency New
AI Summary
Printer on 3rd floor offline since 9am, blocking production reports. Needs immediate triage — escalate to hardware queue.
Extracted by claude-sonnet-4-6 · 2s · 98ms
Add Note
Assign Queue →
Capabilities

Built for how helpdesks actually work

Every feature is designed around the real rhythm of a support team — fast triage, clear context, and zero dropped balls.

🧠
AI-Powered Triage Live
Swap in any LLM — Claude, GPT-4o, Gemini, a local Ollama model, whatever — and it extracts caller name, company, callback number, intent, urgency, and a plain-English summary. No forms. No lock-in. Structured work items, automatically.
📝
Internal Notes
Add team context to any call. Notes stay attached to the record so the next tech has full history before picking up the phone.
📅
MCP Calendar Integration Soon
ComFlow will connect to your calendar via MCP — scheduling callbacks, booking onsite visits, and sending confirmations without a human touching a calendar app.
📋
SOP Knowledge Base Soon
Link your standard operating procedures. The AI will match incoming issues to known workflows and pre-populate resolution steps for the technician.
🔀
Smart Queue Routing Soon
Intent + urgency + workload = the right tech, right now. Automated assignment rules keep the inbox from becoming a pile.
🔌
Engine Interoperability Core Design
ComFlow is built around provider interfaces, not hard dependencies. Swap LLM, STT, and TTS engines at config time. Run them head-to-head. Your infra, your vendors, your call.
LLM · Claude
LLM · GPT-4o
LLM · Gemini
LLM · Local
STT · Whisper
STT · Deepgram
TTS · ElevenLabs
TTS · Kokoro
📊
Provider Metricing Soon
Every extraction run is timed and scored. Compare engines on latency, accuracy, and cost — per call, per queue, over time.
claude-sonnet-4-698ms
gpt-4o-mini142ms
gemini-flash211ms
llama3-local380ms
⚡ Available Today
Start with voicemail.
Grow into AI.
The foundation is live: capture inbound calls, transcribe with your chosen STT engine, extract structured metadata with your chosen LLM, and manage everything from a clean review UI. The AI secretary — engine-agnostic by design — is built on top of this.
Get Started →
Process

Three steps. Zero dropped calls.

ComFlow sits between your phone system and your team, making sure every inbound call becomes an actionable work item.

1
Call comes in
Your telephony system fires a webhook. ComFlow captures the call record, stores the recording reference, and queues it for processing. Nothing is missed.
→
2
AI extracts context
The recording is transcribed by your configured STT engine. Your chosen LLM then pulls out caller name, company, callback number, intent category, urgency level, and a plain summary — all measured and logged.
→
3
Team reviews & resolves
The call lands in the inbox with full context. Technicians review, confirm or correct the AI output, add notes, assign to a queue, and mark resolved — all from one screen.
Roadmap
Where we're
going.

ComFlow starts as a smarter voicemail inbox and grows into a fully autonomous helpdesk co-worker. Each layer builds on the last.

  • 📞
    Voicemail capture + transcription Inbound webhooks, recording storage, STT pipeline
    Done
  • 🧠
    AI metadata extraction Pluggable LLM extraction — Claude, GPT-4o, Gemini, local models
    Done
  • 🖥️
    Human-review inbox UI Review, edit, assign, and resolve calls from a web app
    Done
  • 📡
    Real telephony adapter Twilio / Telnyx integration replacing the fake provider
    Building
  • 📊
    Engine benchmarking dashboard Per-call latency, accuracy, and cost metrics across all provider slots
    Building
  • 🔌
    Real STT / TTS adapters Whisper, Deepgram, ElevenLabs, Kokoro — swappable at config time
    Planned
  • 📅
    MCP calendar integration AI schedules callbacks and visits via Model Context Protocol
    Planned
  • 📋
    SOP knowledge base Match issues to known workflows, pre-populate resolution steps
    Planned
  • 🔀
    Automated queue routing Intent + urgency + availability → right tech, right now
    Planned
  • 🤖
    Autonomous helpdesk resolution AI executes known SOPs end-to-end for common issue types
    Planned
📬

Voicemail only — for now.

Right now, ComFlow is a really smart voicemail system. It captures your calls, transcribes them, and extracts what matters — so nothing falls through the cracks. The AI secretary that books your appointments, walks SOPs, and resolves tickets? It's coming. This is its foundation. Get in early.

Star on GitHub →