Microsoft Open-Sources Agensh: A Multi-Agent Harness for 1,024 Agents
Microsoft Research releases Agensh, a decentralized multi-agent harness enabling up to 1,024 AI agents to collaborate under a 6-hour budget.
Stock photo for illustration only, not from the actual event
- Microsoft Research open-sourced Agensh, a decentralized multi-agent harness.
- Supports seamless scaling from 1 to 1,024 agents on ProgramBench.
- Utilizes a shared Git repo, chat system, and findings board for coordination.
- Operates under a strict 6-hour time budget constraint.
Microsoft Research has officially open-sourced Agensh, a novel multi-agent harness designed without any central orchestrator. The framework enables self-organizing worker agents to coordinate their tasks autonomously through shared infrastructure.
The system allows agents to communicate and share data effectively by leveraging a shared Git repository, alongside integrated chat channels and a dedicated findings board. This setup facilitates continuous insight exchange and status updates without relying on a hierarchical central controller.
Tested on the ProgramBench benchmark, Agensh demonstrated the ability to scale smoothly from a single agent up to 1,024 concurrent agents within a tightly enforced 6-hour budget, highlighting its capability to handle highly complex, resource-intensive collaborative tasks.
Stock photo for illustration only, not from the actual event
The release of Agensh highlights the broader industry shift toward decentralized multi-agent systems rather than relying on heavy monolithic models or rigid hierarchical controllers. By utilizing standard developer tools like Git for coordination, Microsoft addresses traditional communication bottlenecks and scaling limitations inherent in massive agent networks.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment