What this page is
A short orientation note for the local Pi extension and how it supports the autoresearch loop.
Core idea
The project uses a local extension layer to expose research-specific tooling to the agent. That includes experiment execution, result logging, and coordination with the benchmark substrate.
Why it is separated from the main graph
This is part of the tooling substrate around the project, not part of the LLM/content research graph itself.
Repository role
- extension code lives in the local project/vendor area
- it orchestrates the benchmark loop
- it supports the local protocol documented in the research protocol