Common questions about MuxAgent
Everything you need to know about installing, securing, and operating MuxAgent — from first setup to multi-machine workflows.
Getting Started
What is MuxAgent and what problem does it solve?
MuxAgent has two core parts: a graph-based orchestration workflow for agent tasks, and remote control for agent sessions running on real machines. The workflow side structures work into explicit stages such as planning, review, approval, implementation, and verification. The remote-control side lets you monitor and steer those sessions from your phone, even when the machine is elsewhere. Together, they solve the operational problem of keeping agent work moving across interruptions, long-running sessions, and multiple machines.
How do I install MuxAgent?
On macOS or Linux, run the install script: curl -fsSL https://raw.githubusercontent.com/LaLanMo/muxagent-cli/main/install.sh | sh. On Windows, download the latest release from the GitHub releases page. After installing the CLI, start the daemon with muxagent daemon start and pair with the mobile app by scanning the QR code. The full walkthrough is in our getting started guide.
Do I need to create an account to use MuxAgent?
No. MuxAgent does not require a centralized account or subscription. Authentication is device-to-device: you pair your phone with your machines by scanning a QR code. Cryptographic keys are generated locally on your devices. The relay routes encrypted traffic between them but never has access to your credentials or identity.
Security & Privacy
Is my source code visible to MuxAgent's servers?
No. All communication between your phone and your machines is end-to-end encrypted using X25519 key exchange and ChaCha20-Poly1305 AEAD encryption. The relay server routes encrypted traffic but cannot decrypt it — it has no access to your source code, agent output, prompts, or session content. This is a zero-knowledge architecture by design, not by policy. See our encryption deep-dive for the full technical explanation.
Can I self-host the relay server?
Yes. The relay server and CLI are open source. You can run the entire stack on your own infrastructure, which means zero third-party involvement in your agent communication. Self-hosting is documented in the CLI repository. For most users, the default relay is fine since it cannot read your traffic, but self-hosting is available for teams with strict compliance requirements.
What data does the relay server see?
The relay sees only encrypted blobs and connection metadata: which machine IDs are connected, when they connect and disconnect, and the size of encrypted messages. It cannot see message content, source code, agent output, file paths, or any other payload data. Think of it as an encrypted mail router — it knows the envelope exists but cannot open it.
Compatibility
What AI coding agents does MuxAgent work with?
MuxAgent currently supports Claude Code and OpenAI Codex as agent runtimes. The CLI's configuration system allows you to specify which runtime to use per task. The workflow system (planning, review, approval, verification) is runtime-agnostic — it works the same way regardless of which AI model powers the agent. Runtime entries live in the user-level ~/.muxagent/config.json or project-level .muxagent/config.json.
What platforms does MuxAgent run on?
The CLI runs on macOS (Intel and Apple Silicon) and Linux (x86_64 and ARM64). Windows support is available as a manual download from the releases page. The mobile app is available on Google Play for Android (iOS is coming soon). The relay server runs on any platform that supports Go binaries. The CLI and mobile app communicate through the relay, so the machines can be anywhere — local, cloud, or behind a firewall.
Can I use MuxAgent with multiple machines?
Yes — this is one of MuxAgent's core capabilities. You can pair multiple machines with a single mobile app, and each machine appears as a separate entry in the app. The daemon runs independently on each machine, and you can start, monitor, and control agent sessions on any paired machine from one screen. See our guide on running agent work across multiple machines.
Workflows
What happens if my internet connection drops during a session?
A WebSocket disconnect only drops the encrypted communication channel — it does not terminate your agent session or lose any data. Session metadata is persisted locally on both the CLI (local storage) and the mobile app (SQLite). When the connection restores, the daemon automatically reconnects to the relay, re-establishes the encrypted channel, and resyncs events. Your running agent task continues uninterrupted on the machine.
What are the different workflow configurations?
MuxAgent provides four built-in workflow configs: default (plan → review → approve → implement → verify — full ceremony for important work), autonomous (plan → implement → verify — skips approval for quick, reversible tasks), plan-only (plan only — for analysis and exploration without code changes), and yolo (minimal structure for exploratory work). You choose the config per task, matching the process to the risk level. See our workflow config guide for details.
Can I customize the workflow graph beyond the built-in presets?
Yes. The built-in configurations cover the most common patterns, but the underlying system is a directed graph of workflow stages. Workflow bundles live under ~/.muxagent/taskconfigs, and you can clone and edit them to change the graph, prompts, runtime, iteration limits, or clarification settings. The regular config files (~/.muxagent/config.json and .muxagent/config.json) are for user/project configuration, not for defining the task graph itself. The graph workflows post explains the architecture.
How do approval checkpoints work when I'm away from my computer?
When an agent reaches an approval checkpoint, the event is sent to your mobile app through the encrypted relay. You see the agent's plan, review the proposed approach, and approve or reject it from your phone. The agent waits at the checkpoint until it receives your decision — there is no timeout by default. This means you can start a task before a meeting and approve the plan during a break, without returning to the original terminal.
Pricing & Licensing
Is MuxAgent free?
The MuxAgent CLI is open source and free to use. The relay server is provided as a free hosted service. The mobile app is available on Google Play (iOS is coming soon). MuxAgent itself does not charge for usage — however, the AI agent runtimes (Claude Code, Codex) require their own API keys, and usage of those APIs is billed by their respective providers (Anthropic, OpenAI). MuxAgent is the control plane; the AI model costs are separate.
Where is the source code?
The MuxAgent CLI source code is available on GitHub at github.com/LaLanMo/muxagent-cli. The codebase includes the CLI, daemon, relay client, and the encrypted communication layer. You can inspect the encryption implementation, contribute improvements, or fork the project for custom needs. Open source is a trust decision — we believe the code should be inspectable for a tool that handles your source code.
Still have questions?
Check the blog for in-depth essays on workflow design, encryption, and operational patterns. For bug reports and feature requests, visit the GitHub repository.