One source of truth for AI coding agent configuration. Write your project rules once. Compile them to every agent's native format.
You use multiple AI coding agents on the same project: Claude Code, Cursor, Aider, GitHub Copilot. Each one needs its own config file telling it about your project. You maintain them by hand. They drift. Your agents start giving conflicting advice.
Write your project rules and prompts once in .agentsync/. Run agentsync build. All your agent config files stay perfectly in sync — forever.
Today (alpha):
.agentsync/rules/stack.md → CLAUDE.md (Claude Code)
.agentsync/rules/testing.md → .cursor/rules/*.mdc (Cursor)
Planned for v0.2 (compilers exist as stubs today):
→ .aider.conf.yml + CONVENTIONS.md
→ .github/copilot-instructions.md
→ AGENTS.md
npm install -g agentsync
cd my-project
agentsync init # scaffold .agentsync/ with starter rule + prompt
agentsync build # generate enabled agent configs (Claude Code, Cursor)
git commit -am "Add AgentSync"Alpha note: today
agentsync initis a non-interactive scaffold. An interactive flow that auto-detects and imports existingCLAUDE.md/.cursorrulesis planned for v0.2 — see ROADMAP.md.
Done. Your agents are now in sync.
Every feature must advance all five:
| Pillar | What it means |
|---|---|
| USEFUL | Solves real, daily multi-agent pain |
| EASY | Install → first build in under 60 seconds |
| FRIENDLY | Beautiful CLI, helpful errors, encouraging tone |
| COMPLETE | Every supported agent gets a first-class compiler |
| PRIVATE & SECURE | Local-only, no telemetry, no network, no surprises |
AgentSync runs entirely on your machine.
- Zero network calls in MVP
- Zero telemetry, ever
- No accounts required
- Strict file system allowlist — only reads/writes known paths
- Restricted templating — no template includes, no env/fs/process exposed (note:
.agentsync/content is treated as trusted — see docs/privacy.md) - Built-in secret detection — refuses to build if API keys are found in rules
- Right to leave:
agentsync eject --fullremoves all trace - MIT licensed, fully auditable
Suitable for air-gapped, regulated, and enterprise environments.
See SECURITY.md and docs/privacy.md.
v0.1.0-alpha.0 — early but honest. Currently supports:
| Agent | Status |
|---|---|
| Claude Code | ✅ Fully supported (CLAUDE.md) |
| Cursor | ✅ Fully supported (.cursor/rules/*.mdc) |
| Aider | 🚧 Stub (build warns; planned for v0.2) |
| GitHub Copilot | 🚧 Stub (build warns; planned for v0.2) |
| AGENTS.md | 🚧 Stub (build warns; planned for v0.2) |
See CHANGELOG.md for the full alpha-scope notes and ROADMAP.md for what comes next.
- Getting Started
- Per-agent guides
- Error reference
- Privacy & Security
- Release process (for maintainers)
- Security policy
- Contributing
- Decisions log
- Roadmap
- Changelog
git clone https://github.com/__AGENTSYNC_LAUNCH_PLACEHOLDER__/agentsync.git
cd agentsync
pnpm install
pnpm test
pnpm cli -- --helpMIT — see LICENSE.