Engram
A long-term memory plugin for OpenClaw agents that combines exact keyword search over stored facts with vector-based semantic recall, so an agent remembers what a user told it across sessions.
Engram is a hybrid long-term memory plugin for OpenClaw agents, giving an agent a store of facts that survives between sessions instead of forgetting everything once a conversation ends. It combines a SQLite database with full-text search for exact, structured recall of stored facts with a separate vector database for fuzzy, semantic search over the same memories, and queries both together.
Memories are grouped into categories such as preference, fact, decision and entity, and each one carries a decay class, from permanent down to session-only, so how long a memory is expected to matter affects how its confidence fades over time. Auto-capture and auto-recall hooks can pull relevant memories into a conversation and save new ones automatically, and embeddings are generated through the OpenAI API.
The project's own documentation is careful to note that it is unrelated to a different, similarly named tool that compresses code context for coding agents rather than persisting agent memory.