Monad Documentation

An agent’s identity, memory, permissions, and history normally belong to whatever product runs it. Close the product and the work is gone; change vendors and the team starts over. Monad is where agents live. Monad is an open-source agent team runtime from Monadix: a daemon-first runtime with headless architecture, released under the MIT license and installed as one binary. One long-lived local daemon holds the team’s identity, capabilities, permissions, memory, sessions, collaboration state, approvals, and audit history, and keeps them across client closures, restarts, and a change of runtime under any member. Clients render and control that state; none of them owns it. Monad Mesh is the agent team runtime built on that ownership. It gives every agent a durable identity, binds agents to sessions and projects, and coordinates delegation, observation, approvals, and recovery. Agents from many runtimes can join one mesh: third-party agent providers such as Codex and Claude Code, Agent Client Protocol (ACP) agents, peer daemons you own, and Monad Agent Runtime, Monad’s bundled first-party agent runtime and default member. It installs as one binary and runs as one process on the machine you already use. It needs no cluster, message server, gateway, or object store. Under Monad Agent Runtime, every agent action runs with approvals, OS-level sandboxing, filtered egress, and audit. A member backed by a third-party runtime keeps that runtime’s execution and permission model. Monad records its activity and proxies approval prompts when the provider exposes them. Why Monad explains the product boundary.

​

Monad at a glance

This documentation separates operating the product from building and changing it. Simplified Chinese entry points: 文档首页, 快速上手, 架构总览, and 项目首页.

​

User-facing documentation

User-facing pages explain what Monad does and help you complete an operational task. They do not require knowledge of daemon internals.

​

Build the team: Monad Mesh

Monad Mesh is the agent team runtime and the first thing to learn. It decides who is on the team, what each member may do, and how their work is bound to sessions, observed, and approved.

​

Run the first-party agent: Monad Agent Runtime

Monad Agent Runtime is Monad’s own agent runtime and the default mesh member. It is an attached capability, not the product boundary. A mesh can run without it.

​

Operate the daemon

These guides cover the durable daemon and the clients that control it:

​

Tune trust

​

Developer-facing documentation

Developer-facing pages explain public contracts, shipped architecture, extension points, engineering choices, and product design foundations.

​

Runtime internals

The internals follow the daemon-first product boundary: Build against public protocol and client packages. Do not import daemon implementation details. Prefer documented extension surfaces such as skills, atom packs, Model Context Protocol servers, hooks, agent adapters, and Workplace Experiences.

​

Understand the repository

These public pages explain stable design decisions without prescribing the repository workflow: Repository rules and development practices are intentionally not published through Mintlify. Start with CONTRIBUTING.md, then use the repository-only development documentation. Coding-agent coordination lives separately in the repository-only agent documentation. Package-specific documentation stays next to its implementation. Examples include the Web app docs, sandbox hardening status, and RTK client documentation. Agent instructions are not product documentation. Edit .rulesync/rules/, then run bun run agents:sync. Do not edit generated AGENTS.md or CLAUDE.md files.