Persistent memory for AI agents. Index conversations, build knowledge graphs, recall anything — across sessions.
AI agents forget everything between sessions. Every conversation starts from zero. Context is lost. Decisions are repeated. Your agent doesn't know what happened yesterday — let alone last week.
This isn't a limitation of the model. It's a limitation of the stack.
Engram Core indexes your agent's conversation history into a local vector store. It parses transcripts, chunks them intelligently, embeds them with OpenAI, and stores them in ChromaDB — right on your machine.
Your agent can now search its own memory with natural language. No cloud. No third-party storage. Just persistent, private, semantic recall.
Everything an AI agent needs to remember.
Semantic search across all past conversations using OpenAI embeddings and ChromaDB. Find context by meaning, not just keywords.
Set-and-forget cron job indexes new sessions every 30 minutes. Your agent's memory stays current without manual intervention.
Single install script sets up Python venv, dependencies, ChromaDB, initial index, and cron — in under a minute.
Built specifically for the OpenClaw agent framework. Reads session JSONL files directly. Works as an agent skill out of the box.
Up and running in 60 seconds.
# Clone the repo
git clone https://github.com/engramcore/engram.git memory-engine
cd memory-engine
# Install (creates venv, indexes sessions, sets up cron)
./install.sh
# Search your agent's memory
.venv/bin/python search_memory.py "what was decided about the deployment"
# Or install as an OpenClaw skill from ClaHub
npx clawhub@latest install engram-coreWhere we're headed.
Session parsing, chunking, OpenAI embeddings, ChromaDB vector store, natural language search, auto-indexing cron, and install script.
Fact extraction engine powered by GPT-4o-mini. Structured knowledge graph with entities, relationships, and categories. Deduplication and confidence scoring.
Shared memory across multiple agents. Federated knowledge graphs. Selective memory sharing with access controls. Multi-instance sync.