Engram is a cognitive memory system that learns what matters, forgets what doesn't, and builds a knowledge graph that grows smarter over time. No external APIs. No vendor lock-in. Your hardware, your data.
Every session starts from zero. Your agent has no continuity, no history, no accumulated knowledge.
They're short-term buffers that vanish the moment the conversation ends. That's not remembering -- that's forgetting with extra steps.
Markdown files and vector databases don't forget, prioritize, or connect. They can't tell important memories from noise.
Not a filing cabinet. A cognitive system that strengthens useful memories, lets irrelevant ones fade, and discovers connections on its own.
One Node.js process. One SQLite database. Local embeddings. No external APIs. No vendor lock-in.
Memories strengthen with use and fade when ignored. Based on the algorithm behind 100M+ Anki reviews.
Vector similarity, full-text, personality signals, and graph traversal fused via Reciprocal Rank Fusion.
Auto-linking, community detection, PageRank. Memories aren't flat files -- they're a connected web.
Preferences, values, motivations, identity. Every recall shaped by who your agent is talking to.
One Node.js process. One SQLite database. Local embeddings. No OpenAI key. No cloud bills.
Engram is a real server with a REST API, TypeScript SDK, and CLI. Works everywhere, not just MCP clients.
Long memories broken into self-contained facts. Each independently searchable, all linked to source.
When your agent learns something that conflicts with existing knowledge, Engram catches it.
Agents check before they act. Stored rules return allow/warn/block before destructive operations.
Full conversation episodes as searchable narratives. Ask "what happened last Tuesday?" and get a real answer.
Query what your agent knew at any past moment. Debug decisions, audit context drift.
Markdown, PDFs, chat exports, ZIP archives. Full pipeline from raw documents to searchable memory.
What it actually looks like to use persistent memory.
Your agent pulls context from Engram. It knows the project state, your preferences, what you decided last time, and what's still unresolved.
New architecture choices, deployment configs, bug resolutions -- all stored with importance scoring and auto-linked to related memories.
Engram catches contradictions between new information and existing knowledge. Your agent surfaces them instead of silently overwriting history.
Important memories grow stronger. Irrelevant details lose retrieval strength. The knowledge graph reorganizes around what actually matters.
Not a transcript dump. Not a vector search over flat files. Weighted, prioritized, personality-aware context -- assembled from a living knowledge graph.
TypeScript SDK, CLI, or raw HTTP. Pick your weapon.
Also available via MCP for Claude Desktop, Cursor, Windsurf, and other MCP clients. Setup guide →
Included with Engram. Enable when you're ready.
Blocks dangerous operations before they execute. Your agent checks with Eidolon before doing anything destructive.
Relevant memory context injected into every agent session automatically. No manual retrieval needed.
Secrets never leak into prompts. Eidolon intercepts and sanitizes before your agent sees them.
One command. Full memory. Your hardware.