The AI workspace for scientific research
From literature review and hypotheses to code, experiments and tuning — plan, execute, evolve and trace every result in one sandboxed environment.
Run the Linux release
For glibc-based Linux x86_64 (glibc 2.28+ and compatible libstdc++; not Alpine/musl) with Bubblewrap. Run the commands from the download directory.
Download Linux x86_64chmod +x ./ScienceDiscovery-0.2.0-linux-x86_64
./ScienceDiscovery-0.2.0-linux-x86_64 serveNeed Docker, a different OS or architecture, or source mode? See all install options →
This demo uses preset workflows and sample data, not a live research task. Click nodes, replay, or switch views.
Built for the whole research loop
One local workspace that plans the work, runs it safely, searches for better answers, and remembers where every conclusion came from.
Execution environments & workspaces
Run Python, R and shell in a managed, sandboxed environment. Keep code, data and results in a workspace you can inspect.
Learn more →Idea Tree
Explore candidate research directions, assess their designs in separate steps, and use feedback to refine a plan you can test.
Learn more →RSI for research artifacts
Generate and evaluate variants of a program or other measurable artifact, then check the result on held-out data.
Learn more →ScienceMemory & Reviewer
Trace recorded work and evidence through a graph. When enabled, Reviewer flags findings in artifacts that need a closer look.
Learn more →Specialists
Call focused research roles for literature, code and evaluation, or create your own with the right instructions, Skills and tools.
Learn more →Scientific MCP & Skills
Connect literature, databases and tools through MCP. Reuse research methods for retrieval, analysis and delivery through Skills.
Learn more →Run it on your own machine
A single binary, no cloud account. Start the stack and open the UI in your browser.