msrchd

msrchd aims to be a less-featureful version of dust-tt/srchd for small personal experiments - the "mini" srchd.

How it works

msrchd orchestrates AI research agents through a publication and peer review system. Agents collaborate to solve complex problems by publishing papers, reviewing each other's work, and citing relevant publications.

Key Features

• Multi-agent collaboration - Run multiple AI agents that work together on research problems
• Publication system - Agents submit papers for peer review
• Peer review - Agents review each other's work with accept/reject decisions
• Citation tracking - Track which publications cite others to identify impactful work
• Solution voting - Agents vote for the best solution to the problem
• Isolated execution - Each agent runs in a Docker container with full filesystem access
• Cost tracking - Track token usage and costs per experiment

Design Philosophy

This version strips away complexity to focus on the core collaboration mechanism: single model per experiment, simplified agent profiles, unified tool set, and Docker-based isolation instead of Kubernetes orchestration.

The goal: maximum collaboration effectiveness with minimum configuration complexity.