Greplica

Greplica is an open-source CLI tool built by Autoloops that gives AI coding agents persistent, searchable memory of a repository's engineering decisions, constraints, and workflows. It installs via npm, stores a structured knowledge graph locally in SQLite, and integrates with agents like Claude Code, Codex, Cursor, OpenHands, and others. The project is MIT-licensed and fully free to use.

What It Is

Greplica solves a specific problem: every new coding agent session starts cold, forcing the agent to re-learn context it already discovered in prior sessions. Greplica captures the durable parts of those sessions — components, flows, decisions, constraints, and file anchors — and stores them as linked graph nodes. When a new session begins, the agent queries greplica graph context "<question>" to retrieve a ranked Markdown packet of relevant facts before doing any broad exploration. The result is fewer wasted tokens, faster planning, and more reliable agent behavior on complex tasks.

How the Memory Graph Works

Greplica organizes memory into three node types: components (subsystems and modules), flows (multi-step workflows), and claims (specific decisions, constraints, or gotchas). Nodes are linked by edges and anchored to source files. Retrieval combines keyword search, semantic embeddings, and graph traversal to return ranked, relevant results. The graph is stored in ~/.greplica/graph.db on the local machine with no cloud sync and no telemetry in local mode.

• Bootstrap: greplica-bootstrap initializes memory for a repo on first use
• Update: greplica-update-working-memory saves new learnings at the end of a session
• Visualization: greplica graph view opens an HTML graph view in the browser
• Embeddings: supports local (no API key) or OpenAI embeddings

Deployment Model: Local and Managed

Greplica supports two modes. Local mode keeps everything on the developer's machine — no login, no server, no telemetry. Managed mode (requiring v0.2.0 or later) connects a repository to a shared team memory server at memory.autoloops.ai, allowing contributors on different clones and forks to query the same repository memory. Managed mode uses GitHub's browser device flow for authentication and stores only the repository binding, role cache, hook policy, and runtime session metadata locally.

Managed mode includes role-based access (contributor, memory_admin), organization-level inheritance, reusable invite links, and a GitHub Actions reconciliation workflow that audits Memory PRs against exact default-branch code using GitHub OIDC attestation.

Agent Platform Support

Greplica ships agent-native skills and hooks for a wide range of platforms:

• Claude Code and Codex: skills installed to agent home config; hooks inject context guidance and trigger background memory updates
• Cursor: skills at ~/.cursor/skills; guidance delivered via a project rule at .cursor/rules/greplica.mdc
• OpenHands: repo-local install to .agents/skills/ and .openhands/hooks.json
• GitHub Copilot CLI, OpenCode, Factory Droid, Antigravity: also supported

The recommended setup is a one-shot prompt pasted into the agent, which runs the full installer without manual steps.

Update: Greplica v0.2.1

The latest release is v0.2.1, published July 22, 2026. The v0.2.0 milestone introduced managed shared memory mode. The repository was created in May 2026 and has seen active development, with the last push on July 30, 2026. The GitHub README includes benchmark data from SWE-chat held-out planning tasks: the project claims token savings of 26–75% and time savings of 38% or more in showcased runs, with the strongest case cutting tokens by 75% and finishing ~38% faster. These figures are vendor-published benchmark results from the Autoloops team.