NanoFlow-io / engram

What is engram?

Engram is a memory plugin for OpenClaw agents that gives them persistent recall across sessions. It solves the classic problem of AI assistants forgetting everything between conversations by storing facts in two complementary backends: SQLite with FTS5 for fast, exact structured lookups, and LanceDB for semantic vector search when you need fuzzy, meaning-based recall. Every memory gets written to both stores and results are merged intelligently. The plugin exposes tools like memory_recall, memory_store, and memory_forget that agents can call naturally. It runs locally, uses OpenAI embeddings, and includes decay classes so old or low-confidence memories fade away automatically.

Why is it gaining traction?

The hybrid approach is the hook. Most memory systems force you to choose between exact keyword matching and semantic similarity. Engram does both, which matters when an agent needs to recall "my API key is stored in env" (exact) versus "what did I decide about the database stack?" (semantic). The decay system is clever too: permanent facts like birthdays stay forever, while session-level notes expire in hours. This github hybrid rag approach means you get the precision of structured storage with the flexibility of vector search in one package. The CLI is practical for debugging and maintenance without touching the database directly.

Who should use this?

Developers building OpenClaw agents who need their assistants to remember user preferences, project decisions, and context across sessions. If you've been manually re-explaining your codebase or coding style to every new agent session, this eliminates that friction. Teams using AI assistants for complex, multi-session work will benefit most. Early adopters comfortable with new projects will get the most value, since the ecosystem is still forming.

Verdict

A well-designed concept with a credibility score of 0.85% and only 29 stars, this is a young project that shows promise but lacks community validation and battle-testing. The TypeScript implementation is clean and the feature set is thoughtful, but test coverage and documentation depth are unclear. Worth evaluating for OpenClaw-based workflows, but wait for more maturity unless you enjoy bleeding-edge tooling.