Run Claude Code, Codex CLI, and Gemini CLI side by side. See every tool call, catch regressions mid-task, and steer your agents in real time. One workspace for all your locally running agents — lightweight enough to stay out of their way.
Early alpha. Features may be incomplete or unstable. AI agents generate content that may be inaccurate — always review outputs. Report issues
For decades, software meant a human typing code into an editor. That model is ending. The new mode is agentic engineering.
Vibe coding proved you can build apps by directing an agent instead of writing code yourself. Supervibe coding is the next step: you direct groups of agents that collaborate, consult each other, and work in parallel. One agent builds the API while another writes the frontend and a third runs tests. They share context through interpane communication. You watch, steer, and intervene only when needed.
Give each agent a role — API, frontend, tests, review
Watch agents collaborate and catch conflicts in real time
Redirect when agents drift — let the team self-correct when they don't
Multi-provider agent panes, interpane communication, regression detection, and a subagent watcher — built on a Rust backend light enough to run alongside 10+ agents on modest hardware.
Native pane type for Claude, Codex, and Gemini. Structured view of tool calls, reasoning steps, and file diffs. Not a terminal wrapper - a purpose-built agent UI with streaming parser, OAuth flow, and state machine.
Every tool call streamed in real time. Visual diff overlay shows file changes as they happen. Regression detection highlights when an agent reverts correct work. One-click interrupt to redirect mid-task.
Panes talk to each other via Tokio channels. An agent's output streams into another pane's input. Build reactive pipelines - agent A's result triggers agent B. Visual connection indicators show active data flows.
Monitor sub-agents spawned by your primary agent. JSONL stream parsing surfaces every subagent's activity in a dedicated pane view. See multi-level agent hierarchies in real time.
A persistent agent configuration manager. Define working directory, Soul identity, AgentMD, MCP config, skills, and environment variables per agent. SQLite-backed with auto-start, crash-restart, and one-click launch for any provider.
Your agents run on your machine. No cloud sandbox, no waiting for results. Full visibility into every tool call, file write, and network request as it happens. Your data never leaves your machine.
Your agents configure the app for you. They spawn panes, set titles, build dashboards, rearrange the workspace, and update their own status — all through a typed API with sub-millisecond IPC to the Rust backend. You describe the layout you want; your agents make it happen.
Dedicated sysinfo pane with real-time CPU, memory, and network graphs via Recharts. Live status bar with uptime counter, per-pane stats, and agent activity indicators.
Syntax-highlighted code preview pane. Authentic terminal emulation via xterm.js + portable-pty. Per-pane zoom, font size, transparency, and theme. Shell integration across bash, zsh, fish, pwsh.
150-350MB memory footprint — comparable to Warp or VS Code, but backed by a pure Rust engine with zero GC pauses and no heap growth over time. Runs comfortably alongside 10+ agent processes on modest hardware.
Native builds for Windows, macOS, and Linux. Code-signed and notarized on macOS. NSIS installer + portable ZIP on Windows. AppImage + .deb on Linux.
No telemetry, no tracking, no phone-home. Audit every line. Extend it. Ship it. Free forever.
Whether you're running one agent or twenty, AgentMux gives you the visibility to stay in control.
Assign agents to separate research tracks - case law, contracts, regulatory compliance. They verify each other's findings through interpane communication. Full audit trail for malpractice protection.
One agent writes Terraform while another reviews for security and cost. Subagent watcher tracks deploy hierarchies. Sysinfo pane shows live metrics during agent-triggered deploys.
Assign agents to logs, metrics, recent deploys, and config changes. See findings converge on a root cause in real time. The session log becomes the post-incident report.
Spin up agents to explore different approaches in parallel. Have them deliberate, challenge each other's conclusions, and converge on verified results.
See every tool call, file write, and network request an agent makes. One agent scans for vulnerabilities while another reviews the findings. Guardrail observability shows which constraints are firing.
Every tool call, file write, and data access captured with full context. Chain of Thought logging at every handoff. Satisfies EU AI Act, Colorado AI Act, and NIST AI Agent Standards.
Jensen Huang told 30,000 developers at GTC 2026: the agentic AI era is here. NVIDIA projects $1 trillion in demand. Every engineer gets a token budget. The tooling for this shift doesn't exist yet.
Electron-based tools eat 500MB+ of RAM per window. When you're running five agents, that's your entire machine. AgentMux uses 150-350MB of memory - a native Rust binary with no garbage collector, no runtime bloat, and no random freezes. ~152MB portable — no install needed. It's the only tool that won't become the bottleneck.
An agent fixes a bug, then undoes the fix three steps later while working on something else. Without visibility into every decision step, you only discover the damage after it's merged. AgentMux shows you the regression as it happens - with zero rendering lag thanks to async streaming.
Running multiple agents in isolation creates duplicated work, merge conflicts, and cascading errors. AgentMux's interpane reactive communication lets agents share context, coordinate tasks, and respond to each other's output - all through the UI, no custom plumbing required.
Today's tools treat agents as dumb text streams. AgentMux's Agent App API lets agents create panes, render custom views, and modify the workspace layout programmatically. Your agents become first-class UI citizens - they can build the interface they need to show you what matters.
Modern agentic models spawn sub-agents to parallelize work. A single session might fork five subagents for different subtasks. AgentMux's subagent watcher auto-detects spawned sub-agents via JSONL monitoring and opens dedicated pane views. Without this, a subagent can silently regress, corrupt data, or loop indefinitely.
The EU AI Act mandates human oversight. The Colorado AI Act takes effect June 2026. 80% of companies with AI agents report boundary violations. AgentMux captures every tool call with full context and identity - session-level audit exports for regulatory evidence. Runs locally with zero telemetry.
AgentMux was rapidly developed by AI agents running inside AgentMux itself. Multiple agents working in parallel on frontend, backend, and infrastructure - coordinated through the same interpane communication and Agent App API that ships to you. The ultimate proof the tool works: we ship with it every day.
Other tools manage their own agent. AgentMux orchestrates Claude Code, Codex CLI, and Gemini CLI — side by side, with full visibility into every tool call and sub-agent.
| Tool | Runs Your Agents | Sub-agent Visibility | Memory | Price |
|---|---|---|---|---|
| AgentMux | Any provider | Auto-detect | 150-350 MB | Free |
| Claude Cowork | Claude only | Step-by-step | ~200 MB | From $20/mo |
| Codex App | Codex only | Traces dashboard | ~300 MB | From $20/mo |
| Cursor | Cursor only | Composer panel | 400 MB-3 GB | From $20/mo |
| Devin | Devin only | Dashboard | Cloud-only | $20/mo + ACU |
| Warp | Warp + CLIs | Terminal panel | 150-400 MB | Free / $20/mo |
| GitHub Copilot | Copilot only | Session logs | IDE plugin | Free / $10/mo |
AgentMux is building toward a complete agentic operating system: persistent agent configs, multi-agent coordination, and remote access from anywhere.
Phase 1
Define. Store. Launch.
A persistent agent configuration manager built into AgentMux. Define every detail of how an agent starts: working directory, Soul, AgentMD, MCP config, skills, and environment variables. Launch it with one click.
Phase 2
Coordinate fleets of agents.
Multi-agent orchestration built on the unleashd pattern. AgentMux detects swarm worker tags from the live PTY stream, groups sessions by swarm ID, and gives you a real-time SwarmDashboard pane with live status for every worker.
Phase 3
Your agents, from anywhere.
A self-hosted remote access gateway connecting 25+ messaging channels (WhatsApp, Telegram, Slack, Discord, and more) directly to your AgentMux agents via JektRouter. Control long-running agents from your phone while away from your desk. Compatible with existing OpenClaw installs.
Follow development on GitHub
Free and open source. Available for Windows, macOS, and Linux.
Early alpha. Your OS or app store may show a security warning — the binaries are not yet code-signed. This is expected and will be resolved in a future release.
v0.32.90
Or build from source on GitHub