Open Autonomous Generalist Scientist
An open-source framework for fully autonomous scientific research — from literature review to manuscript writing.
OpenAGS orchestrates a team of AI agents that collaborate across the full research lifecycle — literature review, hypothesis generation, experiments, manuscript writing, and peer review. One framework, end-to-end, fully autonomous.
OpenAGS Desktop — Multi-agent research workspace with integrated LaTeX editor
Autonomous Generalist Scientist — Framework and Vision
| Dependency | Version | Required For |
|---|---|---|
| Node.js | >= 20 | Server & UI |
| pnpm | >= 9 | Package manager |
| TeX Live / BasicTeX | any | LaTeX compilation (optional) |
| Docker | any | Sandboxed experiments (optional) |
| Rust | >= 1.75 | CLI agent (optional, for development) |
git clone https://github.com/openags/OpenAGS.git
cd OpenAGS
pnpm install
pnpm buildDesktop app (Electron window + server):
cd packages/desktop
npx electron-vite devThis starts the server on http://127.0.0.1:19836 and opens an Electron window. On first launch, create an account from the login screen, then create a research project from the dashboard.
Server only (browser mode — no Electron):
pnpm --filter @openags/app dev # → http://127.0.0.1:19836Open http://127.0.0.1:19836 in your browser.
Production build:
pnpm build
cd packages/app && node dist/index.js # → http://127.0.0.1:19836┌────────────────────────────────────────────────────────────────┐
│ React UI (browser + Electron) │
│ Chat │ Terminal (xterm.js) │ Manuscript Editor │ Settings │
└──────────────────────┬─────────────────────────────────────────┘
│ WebSocket + HTTP
┌──────────────────────▼─────────────────────────────────────────┐
│ Node.js Server (@openags/app) │
│ /chat → Claude SDK, Codex SDK, Cursor CLI, Gemini CLI │
│ /shell → PTY Terminal (node-pty) │
│ /workflow → Workflow Orchestrator │
│ /api/* → REST API (projects, research, config, skills) │
└──────────────────────┬─────────────────────────────────────────┘
│
┌──────────────────────▼─────────────────────────────────────────┐
│ External Services │
│ LLM APIs │ arXiv │ Semantic Scholar │ Docker │ SSH │ OS │
└────────────────────────────────────────────────────────────────┘
OpenAGS/
│
├── packages/
│ ├── app/ # @openags/app — Application server
│ │ ├── src/
│ │ │ ├── index.ts # Entry point
│ │ │ ├── server.ts # Express + WebSocket server
│ │ │ ├── schemas.ts # Zod schemas (data validation)
│ │ │ ├── config.ts # YAML config loading
│ │ │ ├── errors.ts # Error class hierarchy
│ │ │ ├── providers/ # CLI agent integrations
│ │ │ │ ├── claude-sdk.ts # @anthropic-ai/claude-agent-sdk
│ │ │ │ ├── codex-sdk.ts # @openai/codex-sdk
│ │ │ │ ├── cursor-cli.ts # subprocess + stream-json
│ │ │ │ └── gemini-cli.ts # subprocess + stream-json
│ │ │ ├── research/ # Research tools
│ │ │ │ ├── project.ts # Project CRUD
│ │ │ │ ├── experiment.ts # Docker sandbox (dockerode)
│ │ │ │ ├── ssh.ts # SSH execution (ssh2)
│ │ │ │ └── tools/ # arXiv, Semantic Scholar, citations
│ │ │ ├── routes/ # REST API endpoints
│ │ │ ├── workflow/ # Workflow orchestration
│ │ │ └── messaging/ # Telegram, Discord, Feishu
│ │ └── package.json
│ │
│ └── desktop/ # @openags/desktop — Electron + React UI
│ ├── src/
│ │ ├── main/ # Electron shell
│ │ ├── renderer/ # React SPA
│ │ └── preload/
│ └── package.json
│
├── cli/ # openags-cli (Rust, future)
│ ├── Cargo.toml
│ └── src/main.rs
│
├── skills/ # Skill definitions (SKILL.md format)
│ ├── search-papers/SKILL.md
│ ├── verify-citations/SKILL.md
│ └── agents/ # Agent SOUL.md templates
│
├── docs/ # Documentation
├── pnpm-workspace.yaml # Monorepo workspace config
├── turbo.json # Turborepo build config
└── package.json # Root workspace
Stored at ~/.openags/config.yaml:
# Server settings
workspace_dir: ~/.openags/projects
log_level: info
# API keys (for direct LLM access)
anthropic_api_key: sk-ant-xxx
openai_api_key: sk-xxx
gemini_api_key: xxx
# Experiment sandbox
experiment_sandbox: docker # local | docker | remote
# Remote servers (for GPU experiments)
remote_servers:
- name: gpu-server
host: 10.0.1.50
user: research
key_file: ~/.ssh/id_rsa
gpus: [0, 1, 2, 3]
# Messaging notifications
telegram:
bot_token: xxx
chat_id: xxx
discord:
webhook_url: https://discord.com/api/webhooks/xxxAll settings are also configurable from the UI (Settings page).
CLI Agent Backends
| Backend | Integration | Session Resume |
|---|---|---|
| Claude Code | @anthropic-ai/claude-agent-sdk |
--resume sessionId |
| Codex | @openai/codex-sdk |
codex resume sessionId |
| Cursor | subprocess + stream-json |
--resume=sessionId |
| Gemini CLI | subprocess + stream-json |
--resume cliSessionId |
# Install dependencies
pnpm install
# Development mode
pnpm --filter @openags/app dev # Server only (http://127.0.0.1:19836)
cd packages/desktop && npx electron-vite dev # Desktop app (Electron + React)
# Build all packages
pnpm build
# Lint
pnpm lint
# Type check
pnpm typecheck
# Run tests
pnpm testcd cli
cargo build --release
# Binary at: target/release/openagsIf you use OpenAGS in your research, please cite:
@article{zhang2025scaling,
title = {Scaling Laws in Scientific Discovery with AI and Robot Scientists},
author = {Zhang, Pengsong and Zhang, Heng and Xu, Huazhe and Xu, Renjun and
Wang, Zhenting and Wang, Cong and Garg, Animesh and Li, Zhibin and
Ajoudani, Arash and Liu, Xinyu},
journal = {arXiv preprint arXiv:2503.22444},
year = {2025}
}
@article{zhangautonomous,
title = {Autonomous Generalist Scientist: Towards and Beyond Human-Level
Scientific Research with Agentic and Embodied AI and Robots},
author = {Zhang, Pengsong and Zhang, Heng and Xu, Huazhe and Xu, Renjun and
Wang, Zhenting and Wang, Cong and Garg, Animesh and Li, Zhibin and
Liu, Xinyu and Ajoudani, Arash},
journal = {ResearchGate preprint RG.2.2.35148.01923},
year = {2024}
}