Open source mission control

Coordinate AI agents across every machine you own.

Agent Commander discovers real tmux sessions, streams them live into the browser, and gives you one place to orchestrate approvals, errors, and multi-host activity.

Read the docs GitHub
tmux-native Live streaming Approvals + orchestration Multi-host
Agent Commander dashboard overview

Feature deep dive

Everything Agent Commander does, distilled from the docs.

Session control

Discover tmux panes, spawn new sessions, fork, rename, archive, and bulk-manage workstreams.

  • Session templates for multi-agent workflows
  • Groups, links, and session relationships
  • Snapshots and event timelines
Read session docs →

Interactive console

Stream tmux panes in real time with full terminal semantics and secure control handoff.

  • PTY-based streaming (cursor, color, Ctrl+C)
  • Read-only viewers with control requests
  • Idle timeouts and detach handling
Console streaming →

Orchestrator + approvals

Centralize sessions that need attention: approvals, input, or error recovery.

  • Approval queue with hook + keystroke decisions
  • Session idle/wake controls
  • Summaries for fast context
Orchestrator →

Visualizer

Immersive full-screen activity view for monitoring live agent work.

  • Multiple themes
  • Live activity feed
  • Quick command bar
Visualizer →

Alerts + notifications

Browser, audio, toast, and Clawdbot alerts with per-provider filters.

  • Usage thresholds
  • Approval and error alerts
  • Clawdbot routing (Telegram, Slack, Discord)
Alerts →

Analytics + search

Track usage, approval outcomes, and search across sessions and snapshots.

  • Weekly and per-session metrics
  • Full-text search
  • Project history
Search + projects →

MCP + voice

Configure MCP servers per host or session and enable live voice transcription.

  • MCP config sync
  • Deepgram transcription proxy
  • Session-level toggles
MCP →

Security + control

Role-based access, scoped host tokens, and configurable agentd capabilities.

  • Admin/operator/viewer roles
  • Host-level permission toggles
  • JWT-based control plane auth
Security →

How it works

Agent Commander is a four-part system: dashboard, control plane, agentd, and Postgres.

Agent Commander architecture

1. agentd discovers tmux

agentd runs on each host, polls tmux panes, captures snapshots, and streams output.

2. Control plane brokers events

REST + WebSocket APIs persist session state, approvals, and analytics in Postgres.

3. Dashboard orchestrates

The UI subscribes to live updates, provides terminal control, and manages approvals.

4. Optional services

OpenAI summaries, Deepgram transcription, and Clawdbot notifications plug in as needed.

Architecture details Streaming pipeline

Live views

Real screenshots from the running system.

Sessions view Orchestrator Visualizer

Operations

Run the stack privately and keep the docs public.

Private deployment

Keep the app on your internal domain. Use VPN or Cloudflare Access for gating.

Public docs

Serve this feature dive at agentcommander.co and the full docs at docs.agentcommander.co.

Coolify + Cloudflare

Deploy static docs with a single Dockerfile and manage DNS in Cloudflare.

Coolify guide Ops playbook

FAQ

Is there a public demo?

No. The public site is docs only. The live app runs on private infrastructure.

What does it require?

tmux, Postgres, Node 20+, and Go 1.22+. See the getting started guide.

Can I run it on one machine?

Yes. The control plane and dashboard can run together with local Postgres.