Description
Agensh is a cutting-edge, scalable, self-organized multi-agent harness designed to operate without a central orchestrator. This innovative system allows for the efficient scaling of agents, ranging from 1 to 1,024, significantly improving test-pass rates. The platform is particularly effective in environments where concurrent workers need to gather context, claim sub-tasks, share findings, verify, and merge progress asynchronously.
Read More
The core of Agensh's functionality lies in its lightweight agentic organization infrastructure, which includes a shared workspace, a message interface, and shared context. This setup enables emergent self-organized cooperation that expands from peer coordination to multi-worker integration, standardized workflows, and specialized roles as the organization grows.
Agensh has demonstrated its capabilities through various tests, such as building pandoc from scratch under a 6-hour budget without internet access. The results showed a rise in the final test-pass rate from 33.89% for a single agent to 55.06% for 1,024 agents. Additionally, when tested on the five hardest ProgramBench tasks, the mean final test-pass rate increased from 19.31% for one agent to 28.78% for 128 agents.
This tool is ideal for organizations looking to enhance their cooperation and task management capabilities, particularly in complex environments where traditional orchestrators may limit scalability and efficiency.
- CategoryAI Agent Builders
- DeploymentCloud
- PlatformsWeb
Agensh's Core Features
Scalable multi-agent harness
No central orchestrator
Improves test-pass rates
Shared workspace
Message interface
Shared context
Emergent self-organization
Standardized workflows
How to use Agensh?
Configure: Set up the shared workspace and message interface
Use: Deploy agents to gather context and claim sub-tasks
Optimize: Share findings and verify progress asynchronously
Scale: Expand from peer coordination to multi-worker integration
Agensh's Use Cases
- Task Management
- Workflow Optimization
- Scalable Cooperation
- Context Gathering
- Progress Verification
