https://github.com/opencsgs/csgclaw
Your own personal AI team.
https://github.com/opencsgs/csgclaw
Last synced: 11 days ago
JSON representation
Your own personal AI team.
- Host: GitHub
- URL: https://github.com/opencsgs/csgclaw
- Owner: OpenCSGs
- License: apache-2.0
- Created: 2026-03-25T01:56:39.000Z (about 1 month ago)
- Default Branch: main
- Last Pushed: 2026-04-15T02:45:03.000Z (12 days ago)
- Last Synced: 2026-04-15T04:24:04.920Z (12 days ago)
- Language: Go
- Size: 5.14 MB
- Stars: 0
- Watchers: 0
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Agents: AGENTS.md
Awesome Lists containing this project
README
English | 中文
# CSGClaw
> Your Personal AI Team
CSGClaw is a multi-agent collaboration platform built by OpenCSG — designed around one practical question: **once work becomes non-trivial, how do you get a group of AI agents to operate like a team, without the system becoming heavy or painful to set up?**
## Install
**macOS / Linux:**
```bash
curl -fsSL https://raw.githubusercontent.com/OpenCSGs/csgclaw/main/scripts/install.sh | bash
```
The installer downloads a prebuilt release binary and places it on your `PATH`. Prebuilt binaries are available for macOS arm64 and Linux amd64.
**Build from source:**
```bash
export CGO_ENABLED=1
go mod download
(cd third_party/boxlite-go && BOXLITE_SDK_VERSION=v0.7.6 go run ./cmd/setup)
go build ./cmd/csgclaw
```
## Quick Start
```bash
csgclaw onboard --base-url --api-key --models [--reasoning-effort ]
csgclaw serve
```
Open the printed URL (e.g. `http://127.0.0.1:18080/`) in your browser to enter the IM workspace.
For a fresh config, `onboard` creates a single `default` provider and sets `models.default` to `default.`.
## Model Provider Examples
### Remote LLM API
```toml
[server]
listen_addr = "0.0.0.0:18080"
advertise_base_url = "http://127.0.0.1:18080"
access_token = "your_access_token"
[models]
default = "remote.gpt-5.4"
[models.providers.remote]
base_url = "https://api.openai.com/v1"
api_key = "sk-your-api-key"
models = ["gpt-5.4"]
[bootstrap]
manager_image = "ghcr.io/russellluo/picoclaw:2026.4.15.3"
```
### Local Codex via CLIProxyAPI
```toml
[server]
listen_addr = "0.0.0.0:18080"
advertise_base_url = "http://127.0.0.1:18080"
access_token = "your_access_token"
[models]
default = "codex.gpt-5.4"
[models.providers.codex]
base_url = "http://127.0.0.1:8317/v1"
api_key = "local"
models = ["gpt-5.4"]
[bootstrap]
manager_image = "ghcr.io/russellluo/picoclaw:2026.4.15.3"
```
### Worker Override Example
```json
{
"id": "u-reviewer",
"name": "reviewer",
"description": "code review worker",
"profile": "codex.gpt-5.4",
"role": "worker"
}
```
## Features
- **Multi-agent coordination** — work with a team of specialized agents through a single coordination point, not a pile of chat windows
- **One-click install** — prebuilt binaries for macOS arm64 and Linux amd64; up and running in minutes
- **WebUI out of the box** — browser-based workspace available immediately after `csgclaw serve`
- **Multi-channel support** — connect Feishu, WeChat, Matrix, or other channels when needed
- **Isolated execution** — each Worker runs in a secure sandbox with security boundaries enabled by default
- **Role-based Workers** — specialize Workers for frontend, backend, testing, docs, research, and more
## What CSGClaw Is
CSGClaw gives you one **Manager** and a set of specialized **Workers**, so instead of juggling isolated agents, you work through a single coordination point for defining goals, breaking down work, assigning roles, tracking progress, and collecting results.
```text
┌────────────────────────────────────────────────────────────┐
│ CSGClaw │
│ ┌──────────────────────────────────────────────────────┐ │
│ │ Manager — understands goals, plans, coordinates │ │
│ └──────────────────────────────────────────────────────┘ │
│ ↓ ↓ │
│ Worker Alice Worker Bob │
│ frontend backend │
│ │
│ WebUI / Feishu / WeChat / Matrix / other channels │
└────────────────────────────────────────────────────────────┘
↑ you make decisions
```
**Manager** — receives your goals, decomposes tasks, selects Workers, tracks progress, and consolidates results.
**Workers** — role-specific executors (frontend, backend, testing, docs, research…). Specialization keeps context clean and reduces role confusion.
**Sandbox** — Worker execution is isolated via **Boxlite**, providing security boundaries without requiring Docker.
**Interface** — WebUI out of the box; Feishu, WeChat, Matrix, and other channels available as integrations.
## A Typical Workflow
```text
You: Build a web app prototype — landing page, login, and basic admin view.
Manager: Splitting into tasks.
· Alice → landing page & login UI
· Bob → backend APIs & data model
· Carol → integration checks
You: Add GitHub login to the login flow.
Manager: Updating Alice and Bob.
Carol: Login response is missing the user avatar field.
Manager: Bob updates the API first; Alice updates the UI once the field contract is confirmed.
```
The key isn't that multiple agents exist — it's that **their collaboration is organized**.
## Design Principles
**A lighter Manager built on PicoClaw.**
Most orchestration layers are built for scale. For individuals and small teams running locally, that weight is a liability. PicoClaw keeps the Manager fast to start and cheap to run — without sacrificing coordination capability.
**A lighter sandbox built on Boxlite, not Docker.**
Isolation is non-negotiable, but Docker is overkill for local-first use. Boxlite gives Workers meaningful security boundaries without asking users to install and manage a container runtime. Safety should not come bundled with unnecessary setup burden.
**WebUI first, channel-agnostic by design.**
Many multi-agent systems are tightly coupled to one messaging protocol. CSGClaw ships with a built-in WebUI so you can start immediately, while keeping other channels (Feishu, WeChat, Matrix) as optional integrations — not assumptions.
## Who It Is For
- Independent developers who want an AI team, not just a single assistant
- Small teams that want lower-friction multi-agent collaboration
- Users who value fast startup, lighter runtime, and sensible defaults
## Acknowledgement
CSGClaw is informed by ideas explored in HiClaw around multi-agent usability, while placing stronger emphasis on lightweight runtime, easier local startup, and a platform model not bound to a single communication channel.
## License
CSGClaw is licensed under the Apache License 2.0. See [LICENSE](LICENSE) for details.