Building a minimal AI agent harness that is easily extensible with third-party tools and harness components. Written in Go.
To clone this repo and start the agent:
git clone https://github.com/jerryschen31/minimal-agent.git
cd minimal-agent
go run .go run . -h lists all of the flags.
If you want to use a remote-hosted model, make sure you configure API keys for your LLM of choice. If using a local model with Ollama, see config.local.json for an example configuration.
Running thoughts
Running decisions
Running session notes
MVP 1: I planned, coding agent wrote code (Fable 5.1)
MVP 2: I designed and coded several iterations, with a coding agent as thinking partner and pair programmer (Sonnet 5.5)
dea1a4f: MVP 2 code refactored into main codebase
e5ba43a: Added an interactive line editor for a better chat prompt UX and added a debug mode
b6d73eb: I planned and guided an agent to build an autonomous mode MVP - works well. I'll use this MVP code as a guide for the full build out.
Chat mode is built. Currently working on a full build of autonomous mode with support for more tools (built-in and MCP server). Also looking into implementing a built-in tracing harness component, that you could still swap out with a third-party tracing tool if desired.