Agensh
Microsoft research harness for many coding agents working without a coordinator.
- ai agent framework
- Free
Pricing
Free and open source (MIT).
Overview
Agensh is an open-source research harness from Microsoft for running large teams of AI coding agents, up to 1,024 in the published experiments, with no central coordinator. Each agent picks up sub-tasks, shares findings, checks results and merges its work on its own schedule, using Gitea as the shared workspace and Mattermost for messages. The release includes the harness, launch scripts and a guide for reproducing the paper's results, which report higher test-pass rates as more agents are added. It is research code, not a product, and its agents run commands and change code on their own, so run it in an isolated environment.
Key features
- Tested with up to 1,024 agents
- No central orchestrator
- Gitea workspace and Mattermost messaging
- Reproduction guide and paper
Use cases
Alternatives
Last updated 2026-10-25