Mux - Coding Agent Multiplexer
A desktop & browser app for parallel agentic development, running multiple AI agents on local or remote compute.
- Source repo
- coder/xum
- Stars
- ★ 2.1k
- Last updated
- today
- License
- AGPL-3.0
- Primary language
- TypeScript
- FA score
- 52/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platform
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Developers who need to run multiple AI agents in parallel on the same project, each handling distinct features or code review tasks, using isolated workspaces to avoid conflicts.
- Not a fit if
- Teams needing a Windows desktop build; pre-built binaries are macOS and Linux only
- Commercial projects wary of AGPL-3.0 copyleft obligations
- Source review
- 52/100 · Major gaps
What does this agent do, and when should you use it?
Mux is a desktop and browser application for parallel agentic development, enabling developers to plan and execute tasks with multiple AI agents on local or remote compute. It provides isolated workspaces: local directories, Git worktrees, and SSH-based remote execution. Mux supports multiple models including Sonnet, Grok, GPT-5, and Opus, with integration for local LLMs via Ollama and long-tail models via OpenRouter. Its custom agent loop is inspired by Claude Code, featuring Plan/Exec modes, vim inputs, /compact, and novel opportunistic compaction and mode prompts. A VS Code extension allows jumping into Mux workspaces directly from the editor. Rich markdown outputs (mermaid diagrams, LaTeX) are supported. Installation is via pre-built binaries for macOS and Linux, and full documentation is at mux.coder.com.
Mux creates isolated runtime environments (local directory, Git worktree, or SSH server) as agent workspaces and manages multiple agent instances. Users configure models like sonnet-4-*, grok-*, gpt-5-*, opus-4-*, or connect to local LLMs via Ollama and cloud models via OpenRouter. Each agent runs a custom loop with features like Plan/Exec mode, vim keybindings, /compact command, and opportunistic compaction. The UI provides agent status sidebar, git divergence view, costs and token consumption tab, and a context management dialog for compaction controls. Integration with VS Code extension allows seamless transition to Mux workspaces. Markdown rendering supports mermaid diagrams and LaTeX, and server mode offers a responsive UI for mobile.
- Developers who need to run multiple AI agents in parallel on the same project, each handling distinct features or code review tasks, using isolated workspaces to avoid conflicts.
- Teams that want to leverage Git worktrees to have each agent work on a separate branch, while monitoring git divergence centrally to catch potential conflicts early.
- Developers requiring remote execution via SSH, allowing agents to run on a server with remote compute resources.
- Users who prefer local LLMs for privacy or offline use, integrating Ollama to run models locally without cloud dependencies.
- Developers who live in VS Code and want to jump directly into Mux workspaces via the extension, improving workflow efficiency.
- Teams managing many agents and needing to track costs and token usage, using the costs tab and context management to optimize resource allocation.
How do you install or deploy this agent?
Download pre-built binaries from the GitHub releases page (https://github.com/coder/mux/releases) for macOS and Linux. Follow installation instructions at https://mux.coder.com/install.
How do you use this agent?
After installing, launch Mux. Choose a runtime: Local (run in your project directory), Worktree (git worktrees on your local machine), or SSH (remote server) per the docs at https://mux.coder.com/runtime. Configure models (e.g., sonnet-4-*, grok-*, gpt-5-*, opus-4-*) or add local/cloud models via Ollama or OpenRouter (see https://mux.coder.com/config/models). Create agents and assign tasks, using commands like /compact for context management. For VS Code integration, install the extension (docs: https://mux.coder.com/integrations/vscode-extension). Full documentation at https://mux.coder.com.
What are this agent's strengths and limitations?
- Isolated workspaces (local, worktree, SSH) with a central git divergence view, enabling parallel development and conflict monitoring.
- Multi-model support including leading cloud models (Sonnet, Grok, GPT-5, Opus) and flexible options via Ollama (local) and OpenRouter (long-tail).
- VS Code extension for seamless integration, allowing direct jumps into Mux workspaces from the editor.
- Cost and token tracking, context management dialog, and rich markdown rendering (mermaid, LaTeX) for enhanced observability and output quality.
- Pre-built binaries are only available for macOS and Linux; Windows users must build from source or wait for support.
- AGPL-3.0 license may impose copyleft obligations on derivative works, requiring careful compliance review.
- As a newer project, ecosystem and community support may be less mature, and documentation might be limited.
- Requires network connectivity for cloud models and OpenRouter; local LLM usage requires additional Ollama setup.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Mux - Coding Agent Multiplexer This agent | 52 · Major gaps | Desktop appFree + model costs | ★ 2.1k | today | TypeScript | — |
| OpenClaude — Universal Terminal AI Coding Agent | 42 · Major gaps | CLIFree + model costs | ★ 34k | today | TypeScript | Claude Code · OpenAI API · Claude API |
| Yardsort | 70 · Some gaps | Desktop appFree | ★ 11 | 8d ago | Rust | Codex · Claude Code |
| Ordewell | 67 · Some gaps | CLIFree + model costs | ★ 202 | today | TypeScript | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence: Repository includes AGENTS.md (development guide) but lacks detailed permission model or security policy. There is a postinstall script and many dependencies, but least privilege is not explicitly addressed. User confirmation mechanisms are not documented, though UI may have dialogs. Data flow transparency is insufficient; no explanation of data transmission. Sensitive data handling not mentioned. Dependency security lacks vulnerability scanning or lockfile. External effects (e.g., SSH, file system operations) lack explicit permission controls. Rollback mechanism not mentioned. Source attribution: author is Coder, but unverified. Deductions: lack of specific security documentation and permission control details.
Evidence: Project structure is consistent, with tests and build scripts. Many dependencies, but no lockfile or integrity checks. Failure messages not documented. Deductions: no dependency availability guarantees and error handling details.
Evidence: README describes multiple use cases (local, worktree, SSH), supports multiple models and Ollama/OpenRouter. Capability boundaries not explicit, but features are documented. Trigger precision not detailed. Environment fit: supports macOS, Linux, Windows, and Docker. Deductions: no detailed configuration options and limitations.
Evidence: README is well-structured with installation instructions, feature list, screenshots. Naming is stable, version is clear. Examples and FAQ are linked. Known limitations not explicit. License is AGPL-3.0 with LICENSE file. Versioning has package.json and release workflows. Maintenance responsibility: author is Coder, but unverified. Deductions: missing known limitations and detailed changelog.
Evidence: Output usability: rich UI and documentation. Marginal value: provides parallel agent development with unique features. Cost-benefit: no pricing or resource consumption info. Deductions: no actual usage effectiveness or performance data.
Evidence: Claims in README have documentation links, but no specific evidence. Cross-source verification insufficient, relying on single source. Fact and inference not clearly separated. Deductions: lack of verifiable test results or independent verification.
- Unverified publisher identity, proceed with caution.
- Many dependencies without lockfile, supply chain risk.
- Permission model and security policy not explicit, need code review.
- Data flow and sensitive data handling opaque, further investigation needed.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.31.1-nightly.640 Pre-release
-
Release notes excerpt
Highlights - Model updates: GPT-6.1 Sol and OpenAI Ultrafast mode, Claude Sonnet 5.5 and Haiku 5.5, and Anthropic Fast mode. - Native host computer use in the desktop app. - Artifacts with rich previews, versions, annotations, and a cross-workspace shelf; MCP App views in tool cards and Artifacts. - Advisor, plugins, workflows, timeline, tool search, and heartbeats promoted out of Experiments. - Goals for top-level workspaces and sub-agents, workflow evaluation, and provider model discovery. - Opt-in cost tracking for AI calls from bash commands, including SSH and Coder hosts. …View on GitHub ↗ -
Release notes excerpt
This release primarily brings first-class support for Grok 4.7, Opus 5.5, GPT-6-Sol and GPT-6-Luna.View on GitHub ↗