Persistent, searchable engineering memory for AI coding agents. Greplica addresses the inefficiency of AI coding agents re-learning context in new sessions, which wastes tokens and time. It provides a persistent, maintained memory that agents can query before exploring, aiming to save approximately 50% of tokens and 30% of time during planning. The target audience is users of AI coding agents, offering a benefit of more efficient, reliable, and context-aware agent performance.

How It Works

Greplica captures durable learnings from past agent sessions, such as decisions, constraints, workflows, and file anchors, storing them as components, flows, and claims. When a new agent begins a task, it first queries this memory using greplica graph context "<question>". This query returns a concise Markdown packet containing relevant facts, target files, subsystem boundaries, and prior decisions. This allows the agent to start with established knowledge, avoiding repetitive tasks like broad searches or inferring ownership, thereby improving plan reliability and quality. New learnings can be integrated via hooks or explicit updates.

Quick Start & Requirements

Most users should install Greplica by pasting a prompt from https://raw.githubusercontent.com/Autoloops/greplica/refs/heads/main/docs/agent-install-prompt.md into their coding agent within the target repository. This prompt guides a setup questionnaire, installs Greplica with a chosen hook mode, creates initial context, and can optionally import durable learnings from recent sessions. Greplica requires Node.js versions 22-26. To visualize the current memory, run greplica graph view. For ingesting old sessions, greplica-fast-session-bootstrap is available.

Highlighted Details

• Demonstrated token usage reduction of 40-50%, with a peak of 75.0% in one case (Gemini Voyager sync/auth).
• Achieved significant time savings, around 30%, with a peak of 38% faster completion (Gemini Voyager sync/auth).
• Improved agent planning accuracy, particularly when tasks depend on repo-specific constraints, decisions, and subsystem boundaries.
• Benchmarks on the SWE-chat dataset show substantial improvements in token efficiency and task completion scores.

Maintenance & Community

No specific details regarding notable contributors, sponsorships, community channels (like Discord/Slack), or roadmaps were found in the provided README.

Licensing & Compatibility

The project is licensed under the MIT license. This license generally permits commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The README does not explicitly detail limitations. However, Greplica's effectiveness is contingent on the quality and completeness of past agent sessions and the agent's ability to formulate effective queries. The system operates fully locally with no telemetry, which is a privacy feature but implies data is managed entirely by the user.