Show HN: Agentsh – A Decentralized, Scalable Multiagent Harness

Category: ai-ml

Tags: multi-agent, decentralized, llm-orchestration, scalability, docker-compose

Score: 6.0/10 (Innovation: 7, Technical: 7, Documentation: 5, Utility: 5)

Agensh is a decentralized multi-agent harness that lets up to 1,024 agents self-organize to claim subtasks, share findings, and merge progress without a central orchestrator, using Gitea for the shared workspace and Mattermost for messaging. It demonstrates measurable scaling effects on ProgramBench tasks (e.g., Pandoc test-pass rates rising from 33.89% with one agent to 55.06% with 1,024). The approach is interesting but the release is primarily an experiment reproduction kit bound to specific Copilot/GPT configurations.

Target audience: AI researchers and developers working on multi-agent LLM systems

Repository: https://github.com/microsoft/Agensh · Python · MIT · 24 stars

View on Hacker News