ScienceLab

AI-powered research environment for VS Code. Describe a scientific concept and ScienceLab generates interactive HTML simulations, manages research knowledge, and runs autonomous investigation loops — all inside your editor.

Features

Interactive Simulations

Tell the agent what you want to explore and it produces a self-contained HTML simulation rendered live in a side panel. Simulations support runtime parameters, telemetry, and auto-repair when errors occur.

• Create — generate a new simulation from a natural-language description
• Enhance — iteratively improve an existing simulation with new features
• Fix — automatically detect and repair runtime errors
• Fork — run parallel experiments side by side (Cmd+Shift+F)

20+ Scientific Libraries

ScienceLab ships with skill files that teach the agent how to use major scientific libraries. When a prompt matches a library's domain, the relevant skill is injected into context so the agent produces correct, idiomatic code.

Simulations can exchange data between libraries through a shared window.labData protocol with typed containers, an event bus, and converter helpers.

Research Workflow

• Knowledge Base — save, search, and cross-reference findings across sessions
• Research Roadmap — define a mission with phased milestones; the agent tracks progress
• Journal — timestamped research notes linked to simulation artifacts
• Autonomous Loop — the agent generates its own follow-up questions and iterates (up to 10 cycles)
• Deep Research — launch long-running Google Deep Research investigations and poll for results
• Paper Generation — export a research session as a structured academic paper

VS Code Integration

ScienceLab uses native VS Code APIs throughout:

• Sidebar — activity bar icon with tree views for roadmap, knowledge, messages, tasks, timeline, and portfolio
• Chat Participant — @sciencelab in the VS Code chat panel with /research, /simulate, and /knowledge commands
• Language Model Tools — deep research, knowledge search, and simulation creation available to any chat participant
• Notebooks — .sciencelab notebook format with executable research cells
• CodeLens — "Run in ScienceLab" and "Export Simulation" actions on HTML files
• Debug Adapter — pause, resume, step, inspect, and evaluate expressions inside running simulations
• Test Controller — automated simulation tests (renders, doesn't error, converges)
• File System Provider — browse workspace research data via sciencelab:// URIs
• Decorations — knowledge-linked highlights on simulation source files
• Comments — AI-generated annotations on simulation code
• Timeline — chronological view of messages and journal entries
• Multi-Root — portfolio tree for cross-project search, knowledge cross-referencing, and simulation comparison
• Tasks — deep research and paper generation as VS Code tasks
• Status Bar — current model and research status at a glance
• External Services — arXiv search, Wolfram Alpha queries, GitHub Gist sharing

Multi-Model Support

Works with Gemini, Anthropic (Claude), and OpenAI models. API keys are stored securely in VS Code's SecretStorage. Select a model from the status bar or with Cmd+Shift+M.

Getting Started

Prerequisites

• VS Code 1.95+
• At least one LLM API key (Gemini, Anthropic, or OpenAI)

Install

1. Install the .vsix file: Extensions > ... > Install from VSIX...
2. Open the command palette and run ScienceLab: Configure API Keys
3. Enter your API key for at least one provider
4. Run ScienceLab: Open Researcher (Cmd+Shift+L) to launch the main panel

First Simulation

1. Type a prompt like "Simulate a double pendulum with adjustable mass and length"
2. The agent generates an interactive HTML simulation in the viewer
3. Use the parameter sliders to adjust values in real time
4. Click Enhance or type a follow-up to iterate

Autonomous Research

1. Set a research mission in the roadmap panel
2. Enable Auto Loop in settings (sciencelab.autoLoop)
3. The agent generates synthetic questions, runs simulations, and advances the roadmap automatically

Keyboard Shortcuts

Settings

Adding Custom Skills

Drop .md files into the skills/ directory of your research workspace. Each skill file uses YAML frontmatter:

---
name: my-library
description: How to use MyLibrary for X
trigger: keyword
keywords: [mylib, custom-domain]
---

Usage instructions and code examples here...

• trigger: auto — always included in the system prompt
• trigger: keyword — included when the user's prompt matches any keyword
• trigger: manual — only included when explicitly referenced

Architecture

src/
  extension.ts          — activation, commands, webview panels
  backend.ts            — IPC message routing, workspace/data management
  agent/
    orchestrator.ts     — task dispatch, autonomous loop, skill loading
    agent-host.ts       — ObotoAgent wrapper, event bridging
    tool-tree.ts        — simulation create/enhance/fix tool handlers
    skill-manager.ts    — skill file loading and prompt injection
    system-prompt.ts    — orchestrator system prompt builder
    simulation-prompt.ts — simulation generation prompt builder
    constants.ts        — default model IDs
    providers.ts        — LLM client factory
    event-types.ts      — shared TypeScript interfaces
  providers/
    16 VS Code API integration modules
webview-ui/
  src/
    contexts/           — LabStateContext (central state)
    hooks/              — useAgentChat, useModels, useWorkspace, etc.
    components/         — React UI components
    api.js              — IPC fetch wrapper
    vscodeNative.js     — native VS Code dialog/config bridge
skills/                 — library skill files (.md)

License

MIT