Experienced OSS Contributors Wanted

We are looking for experienced open source contributors to help remove LangGraph from the agent harness and replace it with a simpler, more capable orchestration model.

Summary

Most modern agentic harnesses are built around a simple while true loop and a rich deterministic harness, rather than routing model outputs through an explicit state machine.

Why

LangGraph pushes the system toward explicit graph/state-machine orchestration, which is the opposite of the direction we want.

What we want instead is:

• give models more room to reason and choose the next step
• keep the harness deterministic and responsible for execution, safety, and tooling
• simplify the core orchestration loop so it is easy to understand and evolve

Claude Code is a useful reference point here: it gives the model maximum decision latitude inside a strong harness, and concentrates engineering effort on that harness rather than on graph wiring.

Proposed Direction

Replace the current LangGraph-style orchestration with a simpler control loop:

• a core while true agent loop
• deterministic tool execution and guardrails in the harness
• minimal orchestration logic outside the model

Goal

Free up the model to do more reasoning, while the harness does everything else.