Overview
The goal of this project is to understand multi-agent orchestration: how to coordinate multiple AI agents, each with specific tools and roles, to accomplish complex multi-step tasks. Everything runs locally — the inference, the orchestration, and the browser-based control plane. No cloud APIs required.
This is primarily an educational project. I'm using it to learn about agent architectures, tool integration, and workflow design. It is still in very early stages — more an exploration than a working product.
Approach
- Agents communicate with a local Ollama instance for inference
- Browser-based interface for visualizing and controlling workflows
- Extensible tool registry: agents can call tools like file operations, shell commands, and web requests
- Structured message passing between agents for task delegation
Technologies
TypeScript
Python
Ollama
partially vibe coded