Local project preview

Knowledge Harbor

A self-improving agent team for connecting internal tools through a governed agent workspace.

Open GitHub repository Visit project website Watch demo video

Project description

Knowledge Harbor coordinates focused agents, scores the quality of each completed run, and uses reviewer feedback to improve the next attempt. The team built the full workflow during the hackathon, using Actian for the core agent loop and Band for evaluation, communication, or runtime support. The demo includes a visible activity trace so judges can inspect what each agent did and where human approval was required.

Project links

Technical architecture

A coordinator discovers allowlisted MCP tools, delegates work to role-based agents, and stores an auditable receipt for every external action.

Setup instructions

Clone the public repository, install dependencies, copy .env.example to .env.local, add the documented API keys, and run the development command in the README.

Lessons learned

The team found that clear handoffs and fast evaluation loops mattered more than adding more agents. Actian handled orchestration while Band made the result easier to inspect and improve.

Additional screenshots or links

https://modelcontextprotocol.io/docs/getting-started/intro
https://github.com/modelcontextprotocol/servers

Tools used

Actian

Band

Back to project gallery