Repository installer
Builds a release binary into ~/.local/bin and installs the agent skill.
git clone https://github.com/agentpmhq/rogrep
cd rogrep
./scripts/install.shLocal open source Rust
grep your
coding-agent history
rogrep turns sessions already on your machine into searchable conversations, inspectable exchanges, Git and PR trajectories, and deterministic statistics.
terminal ~/src
$ rogrep x "fix the parser"
$ rogrep trajectory --pr 48
$ rogrep stats top --by tokens --since 7d
$ rogrep tuiGet started
One binary. No account.
Release archives
Download Linux or macOS archives from the v0.1.0 release and verify the included SHA-256 checksum.
Manual Cargo install
cargo install --path crates/rogrep
rogrep skill install1
rogrep syncIndex local sessions2
rogrep doctorCheck provider coverage3
rogrep tuiBrowse interactivelyPrivate by construction
Your history stays yours.
rogrep reads provider session stores and writes only rebuildable Tantivy and SQLite data under its own local data directory. There is no daemon, account, telemetry, analytics, or runtime network request.
Privacy and paths ->session files --- normalize --- local index
durable source rebuildable data
durable source rebuildable data
Workflows
Follow the evidence, not your memory.
Search exchanges
rogrep x "fix the parser"A request and every agent action that followed stay together.
Trace branches and PRs
rogrep trajectory --pr 48Rank conversations from captured Git and GitHub operations.
Inspect failures
rogrep tool_status:failed tool_cmd:cargoCombine full text with structured facets.
Measure usage
rogrep stats projectsRepeatable local SQLite reports.
Agents welcome
Give your coding agent a memory.
The bundled skill teaches agents when to search, how to drill down, and how to cite evidence.
Set up an agent$ rogrep skill install
$ rogrep skill showTransparent benchmarks
Measured on a real, isolated corpus.
The July 2026 comparison used the same ~730 MB / ~275-session corpus and throwaway data directories. It measured indexing, no-op refresh, footprint, and latencynot relevance quality.
Full methodology!