Skip to content
eclipse-enclavePublic

About

Sandbox for running AI coding agents autonomously: isolated, network-restricted, host-safe

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

84 stars

Watchers

7 watching

Forks

Latest commit

 

History

118 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Eclipse Enclave

Website · Getting Started · Docs · Contributing

Run AI coding agents in full autonomy without giving them your host. Enclave runs Claude Code, Codex, Copilot, OpenCode, or Pi in an isolated container, and attaches the Theia IDE to one. More tools, such as Amp and Antigravity, are in enclave-extensions.

Anatomy of an Enclave session: the developer, the project worktree, extra mounts, secrets, and a persistent store feed an ephemeral sandbox running the AI coding agent; all sandbox traffic leaves through a filtering gateway that allows or blocks it

  1. See: the agent sees your project worktree and the extra mounts you choose. Secrets that the gateway can insert into requests reach it only as placeholders. The rest of the host stays out of reach.
  2. Run: the agent runs in YOLO mode, without confirmation prompts, in a container per session on Docker or Podman.
  3. Reach: all traffic leaves through a gateway that allows only allowlisted domains and logs what it allows and blocks.
  4. Keep: logins, config, history, and agent memory persist on the host across sessions.

Quick start

1. Check the prerequisites

  • Linux, macOS, or Windows through WSL2
  • Rootful Docker with buildx, or Podman (rootless works)
  • Git 2.26 or newer, with an author and committer identity configured

Details and setup commands: Installation.

2. Install

Download from the latest release:

sudo apt install ./enclave_*_amd64.deb     # Ubuntu 24.04, x86-64
sudo dnf install ./enclave-*.x86_64.rpm    # Fedora, x86-64

Standalone binaries for Linux and macOS (x86-64 and arm64) update themselves with enclave self-update (Updating). See Getting Started for each platform, or Installation for all options, including building from source.

3. Run

Run Enclave as a regular user from a project directory:

cd my-project
enclave

The first interactive run asks which agent to use and saves the answer as "tool" in ~/.config/enclave/config.json. Log in from inside the agent once; the login is saved and reused on later runs.

Command What it does
enclave Start a session with the configured agent
enclave --tool codex Use a different agent for one run
enclave continue Continue the latest session
enclave --background Start a session you can detach from
enclave attach [name] Reattach to a background session
enclave ps List running sessions (--all, --json)
enclave shell Open a shell in the container
enclave network log Show what the agent tried to reach
enclave info Show config and image details

Good to know:

  • A foreground session ends with its terminal. Closing it, or pressing Ctrl-\, stops the container.
  • Without a complete Git author and committer identity, Enclave exits before starting the container; see Git.
  • Enclave refuses to run as root and to mount your home directory or other sensitive paths; see Installation and Mounts.

What you get

Area Summary Docs
Agents Six built-in tools, more as extensions; per-tool images and automatic updates Tools and images
Network Allowlist-based egress through a gateway, audit log, host port bridging Networking
Auth OAuth logins saved per tool, layered secrets files, credentials kept out of the container Authentication
Sessions Named and background sessions, persistent history, caches, and agent memory Sessions and persistence
Configuration Global and per-project config, tool config overrides, skills, and patches Configuration, Tool config
Extensions Install tools, features, and commands from Git, or write your own Extensions
Custom commands Your own enclave <name> verbs, run on the host or in a session Commands
Backends Docker, Podman, and an experimental QEMU microVM backend Backends
Security What the sandbox protects against and what it does not Security
Graphical frontend Web UI and desktop app to start, watch, and answer sessions HomeShell (preview)

All commands and flags: CLI reference. The full index is in docs/README.md.

Contributing

See CONTRIBUTING.md for build instructions, where changes go, and the pull request checklist. The Developer Guide and Architecture cover the codebase in depth.

Project resources: website, Eclipse project page, security policy, code of conduct, license (MIT), and notices.

About

Sandbox for running AI coding agents autonomously: isolated, network-restricted, host-safe

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

84 stars

Watchers

7 watching

Forks

Releases

Packages

Contributors

Languages