Skip to content
AgentsMesh CLI

Open sourceMITCLI + TypeScript API

One config. Every AI coding tool.

Write rules, commands, agents, skills, MCP servers and permissions once. Generate native config for every tool. Let your agents learn from failures with lessons.

Get started Import existing config Supported tools 
brew install samplexbro/agentsmesh/agentsmesh
No Node.js needed
Claude Code, Cursor, GitHub Copilot, Gemini CLI, Codex CLI, Windsurf and more, each connected to one AgentsMesh source, ringed by a lessons loop: capture after failures, recall before the next edit. Claude Code Cursor GitHub Copilot Gemini CLI Codex CLI Windsurf ··· and more Lessons Subsystem recall capture Claude Code, Cursor, GitHub Copilot, Gemini CLI, Codex CLI, Windsurf and more, each connected to one AgentsMesh source, ringed by a lessons loop: capture after failures, recall before the next edit. Claude Code Cursor GitHub Copilot Gemini CLI Codex CLI Windsurf ··· and more Lessons Subsystem recall capture

How it works

Three commands. No drift.

Config lives in one place. Every tool reads the format it expects.

  1. Write once

    One .agentsmesh/ folder holds rules, commands, agents, skills, MCP servers, hooks, ignore and permissions. Existing tool configs import in one step.

    agentsmesh init

  2. Generate everywhere

    Native files for every enabled tool, written where each tool expects them. Round-trip metadata keeps import lossless.

    agentsmesh generate

  3. Stay in sync

    Diff, lint and check catch drift in review and CI. Watch regenerates on save.

    agentsmesh check

What syncs

Everything a coding agent reads

Each canonical file maps to the native format of every enabled tool. Partial support is visible in the matrix, never silent.

Lessons

Agents that learn from your repo

A shared, git-tracked memory. Recall before every edit, capture after every failure. Same mistake, never twice, in any tool.

  1. Recall

    Before an edit or a risky command, the agent asks the memory what applies to this file.

  2. Act

    It does the work with those rules in front of it.

  3. Capture

    After a failure or a review comment, it writes one short rule with a trigger.

Terminal window
agentsmesh init --lessons && agentsmesh generate


# after a failure, the agent captures a rule
agentsmesh lessons add "Normalize CLI display paths to forward slashes" \
  --topic windows-paths --trigger-file "src/cli/**/*.ts"


# before the next edit, it recalls the rule
agentsmesh lessons query --file src/cli/foo.ts
#   -> Normalize CLI display paths to forward slashes

Lessons guide →CLI reference →

Supported tools

Works with every major AI coding tool

Descriptor-driven support plus plugin targets. Native, embedded and partial coverage is spelled out per tool.

Full support matrix →

FAQ

Common questions

Full support matrix →
Which AI coding tools does AgentsMesh support?

Every major CLI agent, IDE integration and cloud agent, from Claude Code and Cursor to GitHub Copilot, Gemini CLI, Codex CLI, Windsurf and Zed. The support matrix lists native, embedded and partial coverage per feature.

How do I keep CLAUDE.md, AGENTS.md and Cursor rules in sync?

Write the rule once in .agentsmesh/rules/, run agentsmesh generate, and each file is written in its native format. agentsmesh check fails CI when a generated file drifts.

Does it work with config I already have?

Yes. agentsmesh init detects existing tool folders and imports them into .agentsmesh/. Import is lossless, so a later export restores the originals.

What are lessons?

A shared, git-tracked memory of short rules with triggers. Agents recall matching rules before an edit and capture new ones after a failure, in any tool.

Do I need Node.js?

No. Homebrew and the standalone binary run without Node. The npm package needs Node.js 20 or newer and also exposes the TypeScript API.

Is AgentsMesh free?

Yes. It is open source under the MIT license, with no hosted service and telemetry off by default.

Go deeper

Reference and guides