INSTALL.md: install instructions written for a coding agent, so "install recall for me: follow <link>" in Claude Code, Codex or Cursor sets everything up. The agent asks only which model writes memories, since that decides where chat text goes.
recall setup --model SPEC --embeddings NAME answers those questions up front, for agents and scripts.
docs/benchmarks: the benchmark results as data, and charts drawn from it.
Fixed
The Claude plan provider runs Claude Code from an empty folder. From the shared temp folder, Claude Code indexed thousands of files at the start of every call.