rogrep

01Install rogrep

Paste that in a terminal. Nothing runs as root, no daemon is installed, and no account is created. Then index what is already on disk:

rogrep is open source under Apache-2.0. Read about other installation options.

02What does rogrep do?

Every coding agent writes its session history to disk in its own format. rogrep reads all of them, normalizes them into conversations, exchanges and turns, and makes the whole corpus searchable — so you can ask the history before you rebuild the work.

A match returns the whole exchange — the request and everything the agent did in response — not the single line that happened to contain your words. Known key:value facets narrow the evidence; unknown colon tokens stay searchable text.

03Where rogrep keeps things

rogrep won’t write outside its own data directory. It reads the session stores each provider already maintains, and stores only derived search and statistics data in ~/.local/share/rogrep — overridable with ROGREP_DATA_DIR. Source rollouts stay the durable truth, so anything derived can be deleted and rebuilt.

04Trajectories and statistics

Because rogrep captures Git and GitHub operations as part of the record, it can rank which conversation actually produced a branch, commit or pull request — prioritizing the conversation that created it, ranking matching Git operations above text-only mentions, then printing the exact rogrep show command to inspect the turn.

Statistics come from SQLite over the same normalized record, so they are repeatable and computed without asking a model to interpret anything.

05Speed, measured

Every command auto-syncs before it answers. That is only viable because a no-op sweep costs 16ms — which is the reason these numbers matter more than the cold-index figure.