https://github.com/data-fair/tool-ai-sandbox
Minimal sandboxing for AI coding agents in the data-fair stack.
https://github.com/data-fair/tool-ai-sandbox
Last synced: about 1 month ago
JSON representation
Minimal sandboxing for AI coding agents in the data-fair stack.
- Host: GitHub
- URL: https://github.com/data-fair/tool-ai-sandbox
- Owner: data-fair
- License: agpl-3.0
- Created: 2026-03-17T08:03:30.000Z (5 months ago)
- Default Branch: main
- Last Pushed: 2026-03-19T07:48:15.000Z (5 months ago)
- Last Synced: 2026-04-25T21:12:54.801Z (3 months ago)
- Language: JavaScript
- Size: 24.4 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Agents: AGENTS.md
Awesome Lists containing this project
README
# @data-fair/tool-ai-sandbox
Run AI coding tools (Claude Code, OpenCode) inside a Podman-based sandbox with network access but filesystem isolation.
## Install
```bash
npm i -g @data-fair/tool-ai-sandbox
```
Requires [Podman](https://podman.io/) on the host.
## Usage
```bash
# Run Claude Code in the current directory
df-ai-sandbox claude
# Run OpenCode
df-ai-sandbox opencode
# Open a shell in the sandbox
df-ai-sandbox shell
```
Use `-e VAR` (repeatable) to forward environment variables to the container.
## How it works
The CLI pulls a pre-built container image and runs your chosen tool inside it with:
- The current directory mounted as `/workspace`
- Host git config and tool state directories bind-mounted
- `--userns=keep-id` for correct file ownership
- `--network=host` for API access
## Contributing
See [CONTRIBUTING.md](CONTRIBUTING.md).
## License
[AGPL-3.0-only](LICENSE)