Snapshot, redact, encrypt, and restore Claude, Cursor, Codex, Copilot, and VS Code setup from a Git-backed vault.
AgentSync is a Bun-based CLI that snapshots AI agent configuration from your machine, encrypts it with age recipients, and backs it up into a Git-backed vault under a per-machine namespace, so you can copy any machine's setup onto another.
It is for people who keep global agent configuration in tools like Claude, Cursor, Codex, Copilot, and VS Code and want one encrypted backup per machine instead of manually copying files between laptops. Each machine pushes into its own machines/<name>/ namespace; bringing config to a new machine is an explicit copy, never a silent overwrite.
Global install with Bun 1.3.9 or later:
bun install -g @chrisleekr/agentsync
agentsync --versionWithout a global install, run via bunx:
bunx --package @chrisleekr/agentsync agentsync --versionThe fastest way in is the interactive TUI. Running agentsync with no
arguments opens a tabbed dashboard that lets you browse the vault, inspect
local agent content per agent, trigger a push, browse other machines and copy
from them, and migrate configuration between agents:
# Open the TUI (or use the explicit alias `agentsync tui`)
agentsyncIn a non-interactive shell (CI, piped output) bare agentsync falls back to
the status text output so scripts are not broken.
The flag-driven CLI is still the canonical scripting surface:
# Initialise a vault and the local machine key
agentsync init --remote git@github.com:<you>/agentsync-vault.git --branch main
# Push local agent configuration into the encrypted vault
agentsync push
# On a new machine, after running init with the same remote, restore another
# machine's config from its vault namespace (use `copy self …` to restore your own)
agentsync copy <other-machine> claude/The full quickstart, command reference, and architecture model live at the documentation site: https://chrisleekr.github.io/agentsync/.
| Command | Why you run it |
|---|---|
(bare) / tui |
Open the interactive TUI: vault browser, per-agent local view, push, browse machines and copy, migrate, and a Config tab to change settings. |
init |
Create the local vault workspace, machine key, config, and initial remote state. |
push |
Snapshot local agent configs, sanitise secrets, encrypt artefacts, and push to Git. |
copy |
Restore an artefact (or subdir) from a machine's vault namespace to local disk (copy self … for your own). |
ls |
List machine namespaces, or the copyable artifact paths in one (key-free discovery). |
status |
Compare local files with the vault and surface drift (any machine via --machine). |
doctor |
Run environment, key, and vault diagnostics. |
key |
Add, list, or remove recipients (revocation), or rotate the local machine key. |
config |
View or change vault config: agents enabled and secret-handling policy. |
skill |
Remove a skill from the vault explicitly. |
migrate |
Translate configuration between agent formats locally. |
destroy |
Wipe the local vault clone (default) or the remote vault contents via a normal commit. Local agent files (~/.claude, ~/.cursor, …) are never touched. |
Full flag tables and caveats: Commands.
The full documentation is hosted at https://chrisleekr.github.io/agentsync/ and lives in docs/:
- Architecture — system model, push and copy pipelines, security boundaries.
- Commands — every subcommand, flag, outcome, and caveat.
- Migrate — translate config between Claude, Cursor, Codex, Copilot, and VS Code.
- Operations — key rotation, troubleshooting catalogue.
- Contributing — develop from source, run the test suite, release discipline, doc ownership.
Clone, verify, and read the contributor guide:
git clone git@github.com:chrisleekr/agentsync.git
cd agentsync
bun install
bun run checkThe contributor workflow, release discipline, and doc ownership all live in Contributing.
MIT.