Skip to content
Open Source · MIT License

Configure once. Sync every AI coding assistant.

Keep AGENTS.md, skills, commands and MCP servers synchronized across Claude, Copilot, Cursor, Gemini, Codex, OpenCode, and more.

npx @dallay/agentsync init

Works with the tools you already use

ClaudeCopilotCursorGeminiCodexOpenCode

Why AgentSync?

How AgentSync works

  1. Define

    Keep instructions, skills, and MCP configuration in .agents/ — the single source of truth.

    .agents/AGENTS.md
  2. Apply

    Run one command. AgentSync materializes native symlinks — no copies, no drift.

    $ agentsync apply
    ✔ Linked: CLAUDE.md -> .agents/AGENTS.md
  3. Synced

    Every assistant reads the same files. Verify the state anytime.

    ✔ OK: CLAUDE.md -> .agents/AGENTS.md
    Status: All good

Get started in under a minute — no signup, no SaaS, just the CLI:

$ npx @dallay/agentsync init
✔ Created: .agents/agentsync.toml
✔ Created: .agents/AGENTS.md
✨ Initialization complete!

$ npx @dallay/agentsync apply
✔ Linked: CLAUDE.md -> .agents/AGENTS.md
✔ Sync complete

$ npx @dallay/agentsync status
✔ OK: CLAUDE.md -> .agents/AGENTS.md
Status: All good

Full reference lives in the Getting Started guide, including one-off runs, binaries, and source builds.

This documentation is the authoritative project reference.

Canonical Source: All agent instructions and skills are maintained in .agents/AGENTS.md. This is the single source of truth—they get synced everywhere!

If you are deciding how your team should handle .gitignore, start with the Gitignore Team Workflows guide. It explains the default managed mode, the committed-destination opt-out mode, and when --no-gitignore is appropriate.

If you use AgentSync from Windows, the Windows Symlink Setup guide covers native prerequisites, when WSL is the lower-friction option, and how to verify symlink health after apply.