mikamikasuki/forest

What it solves

FOREST solves the problem of "long-horizon research drift," where autonomous agents lose track of their original goals or make errors that compound over time during complex scientific projects. Instead of relying on a simple, linear chat history that can become cluttered or unreliable, it manages research as a structured, persistent, and editable graph.

How it works

FOREST represents your research as a graph of connected nodes—including hypotheses, experiments, results, and manuscript sections. You provide a research question and a budget, and the agent uses this graph to plan, execute code, analyze data, and refine its approach. Because the state is stored in this graph rather than just a conversation, you can intervene at any time to edit plans, branch into new directions, or correct assumptions without restarting the entire process. The system treats agent outputs as proposals that must be validated by executed code and evidence before they are finalized into a paper.

Who it’s for

It is designed for researchers and developers who want to automate scientific workflows—from literature review and hypothesis testing to generating figures and writing manuscripts—while maintaining full control and traceability over the research process.

Highlights

• Editable Research Graph: Keeps all ideas, experiments, and results connected and modifiable, preventing the loss of context.
• Evidence-Based Workflow: Ensures scientific claims are tied to actual experimental data and code execution, not just persuasive text.
• Autonomous or Manual Control: Offers multiple modes, allowing you to run the agent autonomously, work in an assisted mode, or manually direct specific tasks.
• Full Manuscript Generation: Automatically compiles experimental results, figures, and tables into a complete, traceable research paper.
• Budget and Constraint Management: Tracks compute, time, and model usage to keep research within defined limits.