columbarium

Build and evaluate personal AI agents for BitGN benchmark tasks.

Goal

Maximize BitGN score with fast, measurable iteration.

Current focus

The main target is the benchmark-facing agent loop. That core implementation is still in progress.

Main areas

• src/: agent logic and supporting code for the benchmark-facing loop.
• src/model_clients/: provider adapters behind one model-client interface.
• notebooks/: exploratory notebooks for API probing and agent-development experiments.
• workspaces/agent-runtime/: Python dev and test shell.
• workspaces/local-llms/: local model runtime and Ollama workflows used to debug agent code without paid API usage.
• docs/: source-of-truth project docs.

Start here

• AGENTS.md
• docs/README.md
• docs/project/index.md