https://github.com/tjamescouch/org
LLM sandbox
https://github.com/tjamescouch/org
ai-agent ai-orchestration bun cli-tool developer-tools gpt-oss llm local-llm multi-agent open-source patch-review sandboxing typescript
Last synced: 4 months ago
JSON representation
LLM sandbox
- Host: GitHub
- URL: https://github.com/tjamescouch/org
- Owner: tjamescouch
- License: mit
- Created: 2025-08-09T05:29:58.000Z (11 months ago)
- Default Branch: main
- Last Pushed: 2026-02-07T16:45:40.000Z (4 months ago)
- Last Synced: 2026-02-08T00:58:40.293Z (4 months ago)
- Topics: ai-agent, ai-orchestration, bun, cli-tool, developer-tools, gpt-oss, llm, local-llm, multi-agent, open-source, patch-review, sandboxing, typescript
- Language: TypeScript
- Homepage:
- Size: 2.45 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 10
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Security: SECURITY.md
Awesome Lists containing this project
README
# org
_Agentic dev helpers that can "do things" in your repo while you watch, review, and approve every change._
- ✏️ **Conversational**: give a task, the agent plans and executes.
- 🤝 **Collaborative**: works as part of a multi-agent setup, coordinating steps and maintaining a conversation between the agents.
- 🧰 **Tools**: `sh`, `apply_patch`, etc. 100% auditable.
- 🔒 **Safe-by-default**: denylist for dangerous paths; *every* change turns into a patch you review before it lands.
- 🧪 **Deterministic runs**: all stdout/stderr, artifacts, and a git patch are saved under `.org/runs/` for inspection or CI upload.
- 🛡️ **Optional containerized sandbox**: run steps inside a Podman container with no network by default (quickstart config allows egress to the host LLM at 192.168.5.2:11434).
> Security note. org is safe-by-default but not a hardened sandbox. For a stronger boundary, run it in the provided VM profile (Lima on Apple Silicon, VirtualBox on Intel). See [SECURITY](SECURITY.md) for hardening.
## Vision
Developers and operators deserve AI teammates they can **trust, understand, and control**. Most tools either go fully automated (opaque, risky) or stay passive (just suggest, never act). This project takes a different stance:
* **Dev + DevOps, one CLI**: A single agentic workflow that works for both coding and operations. Write code, run infra, and patch your repo — all inside the same safe loop.
* **Multi-agent as a first-class feature**: Collaboration isn't a bolt-on. Agents can coordinate, debate, and hand off tasks, while you oversee and approve every change.
* **Sandbox + patch review**: Every command runs in isolation. Every file change is surfaced as a patch. Nothing merges until you say so.
* **Safety as a value, not a feature**: Guardrails, glob-based policies, and human-in-the-loop review make it impossible for an agent to "go rogue" in your repo.
* **Composable workflows**: Use it solo as a powerful dev helper, or spin up multiple agents to cover testing, ops, and review simultaneously.
The vision is simple: **an agent CLI that feels like a real teammate — fast, opinionated, and safe.**
## Quick Start
### macOS (Apple Silicon: M1/M2/M3) — Lima backend
```sh
# 1) Install via Homebrew (CLI + VM helpers)
brew tap tjamescouch/org
brew install org lima socket_vmnet
# 2) One shot: create + provision the VM and attach when ready
orgctl quickstart
```
### Running
```sh
org
You > @@alice write "Hello World" to a file `hello-world.txt`
```
> Tip for contributors who installed an older brew orgctl: prefer the repo copy during development with PATH="$PWD:$PATH" or run ./orgctl ….
When the tool wants to touch the filesystem, it first produces a **patch**. You can read, approve, or reject it. After approval, the agent continues.
All artifacts for each run land in `.org/runs//` with comprehensive logs and the final patch.
[](https://asciinema.org/a/741256)
## Installation
See [INSTALLATION.md](docs/INSTALLATION.md) for complete setup instructions.
## Documentation
- **[Installation Guide](docs/INSTALLATION.md)** - Setup for macOS and Linux not including VM setup.
- **[VM Configuration](docs/VM_CONFIGURATION.md)** — Lima/VirtualBox setup, security model, and troubleshooting.
- **[Configuration](docs/CONFIGURATION.md)** - LLM settings, environment variables, and options
- **[Usage Guide](docs/USAGE.md)** - CLI options, examples, and common workflows
- **[Architecture](docs/ARCHITECTURE.md)** - How it works under the hood
- **[Troubleshooting](docs/TROUBLESHOOTING.md)** - Common issues and solutions
- **[Testing](docs/TESTING.md)** - Running and writing tests
- **[Architectural Review](docs/ARCHITECTURAL_REVIEW.md)** - org according to org
## License
MIT
