Introducing Agent M Mobile & Web 2.0

The Autonomous AI Meeting Companion That Listens, Remembers & Takes Action.

Never manually take notes or lose follow-ups again. Agent M deploys headless browser workers to join your Google Meets or turns your mobile phone into an in-person room recorder — extracting prioritized action items and syncing with your semantic vector knowledge graph.

AC
ML
SD
PK
Join 4,821+ engineers and product leaders on the waitlist
pgvector Cosine Search • 12ms
Action Items Auto-Prioritized
Agent M Android App Live Dashboard
Agent M Command Center — http://localhost:5173
Agent M Web Dashboard Command Center

Built From the Ground Up for Real Engineering Teams

Unlike legacy note-taking bots that annoy attendees and dump unformatted walls of text, Agent M operates silently, understands context deeply, and executes real workflows.

Isolated Headless Docker Workers

Drop any Google Meet URL. The receiver dynamically spawns a short-lived Docker container running stealth Camoufox or Playwright Chromium. It joins silently, parses raw caption streams with speaker deduplication, and destroys itself upon meeting conclusion — zero audio intrusion.

Camoufox Stealth Engine Docker Container SDK WebSocket Ingest Zero Telemetry Leak

In-Person Physical Room Capture

Place your phone on the conference table. Agent M captures room audio with live dynamic waveform visualization, segmenting turns and feeding them into the LLM pipeline instantly.

Ranked Action Prioritization

Action proposals are extracted in real-time with confidence scores. Easily drag to re-order priority ranks (#1, #2, #3), assign owners, and approve or deny tool executions with one tap.

3-Tier pgvector Semantic Memory

Agent M builds a persistent, cross-meeting knowledge graph powered by PostgreSQL, `pgvector`, and FastEmbed embeddings. It separates memory into Entity Graph (user preferences & architecture terms), Temporal (upcoming sprint deadlines & windows), and Meeting highlights.

ENTITY_GRAPH TEMPORAL_MEMORY MEETING_CONTEXT FastEmbed Vectors

Experience Agent M Live in Your Browser

Try out the companion's core capabilities right now — test the live waveform speech pipeline, reorder ranked actions, and query semantic vector memory.

In-Person Mic Recording Stream 44.1 kHz Active
Live Speech Turn Stream (Deduplicated & Normalized)
🎙️ Alex (Lead Dev) Just now
Our receiver API handles both Google Meet worker containers and mobile speech turns seamlessly.
🎙️ Sarah (PM) 20s ago
Let's make sure the Expo Android build includes human approval controls for high priority actions.
Click ▲ Up or ▼ Down to adjust priority rank, or click Approve Action to trigger the connected agent tool.
#1
Setup continuous deployment pipeline for Expo iOS & Android builds.
Assigned: Alex (Lead Dev) • Tool: github_issue_creator
#2
Send physical meeting summary report to #engineering channel.
Assigned: Sarah (PM) • Tool: slack_notifier
#3
Schedule follow-up architecture review session with security team.
Assigned: Michael (Architect) • Tool: calendar_invite
🔍
TEMPORAL 96% Cosine Match

Upcoming receiver API deployment window scheduled for Friday 8 PM UTC.

ENTITY_GRAPH 94% Cosine Match

Mobile client authentication defaults to dev mode bypass with JWT verification for production.

👋 Hi! I'm Agent M. Click any quick prompt above to test how I extract intelligence and answer questions from active meeting context!

Zero-Trust Architecture Designed for Total Privacy

Agent M gives your team complete control over your sensitive meeting transcripts, action items, and intellectual property. Deploy on your own AWS ECS, Google Cloud Run, or on-prem hardware with standard Docker Compose.

  • Isolated Meeting Runtimes: Each meeting gets its own ephemeral container destroyed right upon disconnect.
  • Multi-LLM Flexibility: Switch between Anthropic Claude (3.7 Sonnet / Haiku), Google Gemini 1.5 Pro, or AWS Bedrock with one config variable.
  • No Audio Storage Required: Caption streams are processed in-memory with cumulative deduplication and discarded securely.
📱 Android / iOS / Web Client Client UI
▼ HTTP / WebSocket Event Ingest
🚀 Receiver API Server (FastAPI / asyncpg) Backend
▼ Spawns Dynamic Worker Containers
🤖 Camoufox / Playwright Meet Runtime Docker Worker
▼ Embeddings & Vector Similarity
🧠 PostgreSQL + pgvector + FastEmbed Vector DB

How Agent M Compares to Legacy Meeting Bots

See why modern product engineering teams are switching to Agent M.

Feature / Capability Agent M (Our Agent) Otter.ai Fireflies.ai Fathom
In-Person Mobile Room Recording (Android & iOS) ✓ Yes (Waveforms & Turns) ✓ Basic Audio ✗ Online Only ✗ Desktop Only
Isolated Headless Browser Workers (Zero Audio Lag) ✓ Yes (Camoufox / Docker) ✗ Intrusive SIP Bot ✗ Intrusive Bot ✗ Client Recording
Ranked Action Prioritization (#1, #2, #3) ✓ Interactive Drag & Rank ✗ Unranked Text List ✗ Basic List ✗ Flat Text
pgvector Cross-Meeting Semantic Memory Graph ✓ 3-Tier Vector Graph ✗ Flat Keyword Search ✗ Keyword Index ✗ Per-Meeting Only
Automated Tool Execution with Human Approval ✓ GitHub, Slack, Jira ✗ Read Only ✗ Limited Zapier ✗ Read Only
Self-Hosted Docker / On-Premise Support ✓ 100% Self-Hostable ✗ Cloud SaaS Only ✗ Cloud SaaS Only ✗ Cloud SaaS Only

Loved by Engineering Leads and Product Teams

★★★★★

"The ability to rank action items on my Android phone right after an in-person architecture sync and push them as GitHub issues has cut our meeting overhead by 80%."

AC
Alex Chen
Staff Engineer • CloudScale
★★★★★

"Self-hosting Agent M on our private VPC was completely frictionless. The headless Camoufox workers mean no creepy bots sitting in our Google Meet video grid."

MS
Marcus Sterling
VP of Infrastructure • FinTech Corp
★★★★★

"The vector memory search is insane. Asking 'What did Devon say about the billing flaky test 3 weeks ago?' and getting the exact turn and context in seconds is magic."

SL
Sarah Lin
Principal PM • HyperScale AI

Frequently Asked Questions

We are onboarding teams in weekly batches to ensure optimal LLM performance and dedicated onboarding support. When you join the waitlist, you receive a personal ticket ID and referral link. Sharing your link moves you up 50 positions per referral.
Yes! Agent M provides official Docker Compose definitions and Helm charts. You can run the FastAPI receiver, PostgreSQL/pgvector database, and dynamic worker images entirely within your private cloud (AWS, GCP, Azure, or bare metal) with zero external telemetry.
The companion mobile app utilizes high-fidelity microphone input with on-device speech turn segmentation and live audio waveform visualization. Spoken turns are streamed directly to your receiver API via WebSockets for real-time insight generation.
Agent M has native support for Anthropic Claude (Claude 3.7 Sonnet & Haiku), Google Gemini 1.5 Pro & Flash, as well as AWS Bedrock endpoints. You can select your preferred model per meeting or globally in settings.
Agent M uses a strict human-in-the-loop safety model. When actionable directives are identified during a meeting, they are queued with confidence scores. You review and re-order their priority ranks before approving tool execution to GitHub, Slack, or Jira.

Ready to Experience Truly Autonomous Meetings?

Join the waitlist today to secure priority access to the Android & iOS companion apps and self-hosted server releases.

Join the Alpha Waitlist Now Read Architecture Docs