Professor

collaborate

Trainee

submit

🧑‍🔬

You (Editor)

Senior. Junior. You. The work unit, reimagined.

Autonomous Lab

You did the research. Now you judge it.

AI takes the PI and Trainee roles — designing experiments, running analysis, writing the paper. You become the Editor: the sole decision maker who curates, judges, and steers the science.

See How It Works Browse Characters

📝 LaTeX Papers 🧬 Skill Containers ⏱️ 24-Hour Sessions ⚙️ Fully Configurable 🤖 Multi-Agent

🤖 Multi-Agent Orchestration

Each role gets its own context window — using the CLIs you already have

🔌

Zero Extra Cost

Multi-agent mode spawns separate CLI agents (claude, codex, cursor-agent) for each role. It uses the subscriptions you already pay for — Claude Pro, ChatGPT Plus, Cursor Pro. No API keys needed.

🧠

Fresh Context Per Role

In single-agent mode, one context window fills with mixed PI and Trainee concerns. Multi-agent gives each role a dedicated context, so the PI thinks like a PI and the Trainee codes like a Trainee.

👥

Parallel Trainees

A PI can have multiple trainees working simultaneously. A Data Analyst, a Writer, and a Code Developer — each with their own focus area and context, running in parallel. If one fails, the others continue.

agents:
  pi:
    provider: claude-cli
    model: opus
  trainees:
    - name: "Data Analyst"
      focus: "statistics, figures"
    - name: "Writer"
      focus: "paper, LaTeX"

⚙

One Line to Enable

Add orchestration: multi to .autolab/config.yaml. That’s it. If the CLI isn’t installed, it falls back to single-agent automatically. You can also mix providers — e.g. Claude as PI, Codex as Trainee.

orchestration: multi
agents:
  pi:
    provider: claude-cli
    model: opus
  trainee:
    provider: codex-cli
    model: o3

🤝

Agent Teams

In Claude Code with agent teams enabled, multi-agent mode auto-upgrades to native teams. Trainees become teammates that self-coordinate, message each other, and share a task list. Same config, richer collaboration.

🛡

Graceful Fallback

If a CLI binary is missing, an API key isn’t set, or anything goes wrong — the system automatically falls back to single-agent mode for that turn. No configuration changes, no broken sessions. It just works.

📜 How It Works

Watch a research session unfold, step by step

🔬 Professor

Idle

🧪 Trainee

Idle

Consultants

Dr. Tanaka

Statistician

📜 Meeting Log

Click Start Tutorial to begin

Watch how PI and Trainee collaborate on a research project

Step 0 / 9

📦 Inventory

📁data0

📁scripts0

📁figures0

📁results0

📁paper0

📄 Paper

0%

abstract--

introduction--

methods--

results--

discussion--

🧬 Characters as Skill Containers

Each character is a persona backed by a curated set of scientific skills

Dr. Maria Chen

Computational Biology PI

scanpy pytorch-lightning scvi-tools scientific-writing scientific-visualization statistical-analysis

➡

📄 pi.yaml

title: Computational Biology PI
expertise: single-cell genomics and ML
goal: discover cell-type-specific
  gene regulatory programs
skills:
  - scanpy
  - pytorch-lightning
  - scvi-tools
  - scientific-writing
  - scientific-visualization
  - statistical-analysis
personality:
  - "Visionary: identifies novel
    biological questions"
  - "Rigorous: demands reproducible
    pipelines"

How It Works

1

Define a Persona

Give your character a title, expertise, and personality traits. These shape how the AI agent approaches problems and communicates.

2

Attach Cursor Skills

Each skill name maps to a SKILL.md file—the same agent skills used by Cursor and Claude. The AI reads these instructions when acting as this character. Skills include scanpy, scientific-writing, pytorch-lightning, statistical-analysis, and 200+ more.

3

Deploy to Your Lab

Drop the YAML file into your project's .autolab/profiles/ folder. The character is now part of your research team, with all attached skills active.

🏪 Character Marketplace

Browse community-created characters or share your own

1

Fork the template repo

Start from our template which includes the YAML schema, a skills/ folder with example SKILL.md files, and a README.

Fork Template Repo

2

Add character YAML + skill folders

Each character is a self-contained skill bundle. The skills/ folder contains Cursor-compatible SKILL.md files.

your-character/
├── character.yaml
└── skills/
    ├── my-analysis/
    │   └── SKILL.md
    └── my-writing/
        ├── SKILL.md
        └── scripts/  # optional

3

Submit your repo URL

Open an issue on our main repo with your GitHub repo link. We'll index it automatically.

Submit Your Repo

⭐

Get ranked by stars

Once indexed, your character appears in the marketplace. The more GitHub stars your repo gets, the higher it ranks. Official characters from albert-ying get a verified badge.

🚀 Get Started

Set up your own Autonomous Lab in minutes

1

Install the MCP Server

Add Autonomous Lab to your Cursor MCP settings (~/.cursor/mcp.json):

{
  "mcpServers": {
    "autonomous-lab": {
      "command": "uvx",
      "args": ["autonomous-lab"],
      "timeout": 600,
      "env": {
        "MCP_WEB_PORT": "8766"
      }
    }
  }
}

2

Add Character Skills

Install skills in ~/.cursor/skills/. Each skill is a SKILL.md that the AI reads when a character uses it:

# Skills directory structure:
~/.cursor/skills/
  scientific-skills/
    scanpy/SKILL.md
    scvi-tools/SKILL.md
    pytorch-lightning/SKILL.md
    scientific-writing/SKILL.md
    statistical-analysis/SKILL.md
    ...

# 200+ skills available
# Characters reference these
# by name in their YAML

3

Customize Your Team

Download characters from the Marketplace or create your own in .autolab/profiles/:

title: Your Custom PI
expertise: your domain
goal: what to achieve
skills:  # Cursor skill names
  - scanpy
  - scientific-writing
  - statistical-analysis
personality:
  - "Trait: description"

4

Run the Loop

The PI-Trainee loop runs, then you act as Editor:

# The research cycle:
autolab_next  # PI sets agenda
autolab_next  # Trainee executes
autolab_next  # PI reviews
  ...          # iterate

# When paper is ready:
autolab_editorial # You decide!
# Invite reviewers, then:
# Accept / Minor / Major / Reject