mixpeek/amux

What it solves

amux provides a centralized control plane for managing multiple AI coding agents (such as Claude Code, Codex, Gemini CLI, and Ollama) simultaneously. It solves the problem of coordinating a "fleet" of parallel workers, preventing task duplication, and providing a unified interface for steering, monitoring, and scheduling autonomous agent runs.

How it works

Built as a single Rust binary with a SQLite backend, amux acts as an orchestrator that manages agent sessions (typically via tmux). It uses a set of core primitives to coordinate work:

• Board: A shared kanban board with atomic claiming to ensure no two workers grab the same task.
• Workers: Parallel agent sessions with durable identities and the ability to peek into peer terminals.
• Schedules: Cron-style recurring prompts for autonomous overnight or periodic runs.
• Groups & Scope: Logical organization of workers that share memory, environment variables, and access gates.
• .mdai Files: A system of "computed markdown files" that form a directed acyclic graph (DAG). These files use models to synthesize information from other files or folders into a final output, with built-in caching and cycle detection.
• Steering: A web dashboard or mobile app allows users to send messages or redirect running sessions in real-time.

Who it’s for

It is designed for AI engineers and developers who run multiple AI coding agents in parallel and need a way to coordinate them as a team rather than managing individual, disconnected sessions.

Highlights

• Atomic Task Management: Prevents race conditions when agents claim tasks from a shared board.
• Inter-Worker Communication: Supports origin-stamped messaging between agents.
• Local-First Architecture: Self-hosted, SQLite-backed, and distributed as a single Rust binary.
• Dynamic Model Switching: Ability to swap the model or provider for a running worker without losing context.
• Self-Healing: Includes a watchdog to restart crashed sessions and replay missed messages.
• Computed Markdown (DAG): Allows for complex, multi-step synthesis of documentation and research using .mdai files.