What is Greplica?
Greplica is a persistent, searchable engineering memory layer for AI coding agents. Instead of re-learning your repository from scratch every session, your agent can query Greplica before exploring, so it starts with the decisions, constraints, workflows, and file anchors that earlier sessions already uncovered.
Why it matters
Coding agents often burn tokens and wall-clock time grepping around to rebuild context they already had. Greplica keeps the durable parts of past sessions — components, flows, and claims — and serves them back as a small Markdown packet when the agent asks a question.
How it works
- Past sessions reveal repo-specific decisions, constraints, workflows, and file anchors.
- Greplica stores the durable parts as components, flows, and claims.
- A new agent asks with
greplica graph context "<question>"before broad exploration. - The agent uses facts, target files, subsystem boundaries, and prior decisions.
- Memory updates via hooks or
greplica-update-working-memoryafter work sessions.
Modes
- Local mode stays fully local with no telemetry and no login required.
- Managed mode connects an authorized repository to shared team memory, so contributors on different clones and forks can query the same repository memory.
Benchmarks
On held-out planning tasks built from SWE-chat, Greplica cut token usage by 40–50% in several showcased cases. In the strongest measured run, it used 75.0% fewer tokens and finished about 38% faster.
Getting started
Greplica requires Node.js 22–26. Most users install it by pasting the official agent install prompt into their coding agent from inside the target repo. You can then visualize memory in a browser with greplica graph view.
Supported platforms
Install targets include codex, claude, copilot, cursor, opencode, openhands, factory-droid, and antigravity, with local or OpenAI embeddings.