Your AI's memories
shouldn't belong to OpenAI
Every AI platform locks your agent's memories in their walled garden. Switch providers and start from zero. Engram gives you portable, private, platform-agnostic memory — self-hosted or cloud.
Every AI agent wakes up blank
Your agent doesn't remember yesterday's conversation, last week's decision, or the user's name. Most “memory” solutions bolt vector search onto chat history and call it a day. That's not memory — that's ctrl+F.
And worse — the platforms that do offer memory lock it inside their walled gardens. Your agent's knowledge belongs to them, not you.
How it works
Remember
Your agent captures knowledge from every conversation. Engram extracts facts, preferences, events, and decisions — classifying each by type and scoring importance.
Consolidate
The Dream Cycle runs nightly — deduplicating, scoring, pruning stale memories, and forming connections. Like your brain processing short-term into long-term storage.
Recall
Ensemble search across multiple embedding models finds the right memory in 124ms. Temporal awareness means recent context wins when it matters.
import { Engram } from '@engram/client';
const engram = new Engram({ apiKey: 'ek_...' });
// Remember
await engram.remember("Beaux prefers dark mode");
// Recall
const memories = await engram.recall("UI preferences");
// → [{ content: "Beaux prefers dark mode", score: 0.94 }]Built for production
Not a toy. Not a wrapper around vector search. Real memory infrastructure with the features you need.
Dream Cycle
4-stage nightly consolidation: score refresh, dedup, pattern extraction, staleness pruning
Ensemble Search
Multiple embedding models with Reciprocal Rank Fusion for higher recall accuracy
Memory Health Score
Real-time cognitive health monitoring — freshness, coverage, dedup health, vitality
Safety Detection
16 patterns catch PII, medical info, financial data, and credentials before they persist
Memory Pools
Multi-agent shared memory with granular access control and attribution tracking
Local Embeddings
Run models on your own GPU with engram-embed. Zero API costs, total privacy
Self-host everything
Engram was built self-hosted first. Every feature works on your machine with no cloud account required.
Your data, your machine
Everything runs locally. Your memories never leave your hardware. Zero cloud dependency, total privacy.
Local embeddings on Metal GPU
Run 4 embedding models on Apple Silicon or CUDA. No API costs, no rate limits, no latency penalties.
Running in minutes
Setup wizard walks you through account creation and model configuration. One command to start.
All features unlocked
No plan limits. No artificial restrictions. Dream Cycle, ensemble search, code search, knowledge graph — everything, free.
Start local, add cloud when you need it
Hybrid mode connects your self-hosted instance to OpenEngram Cloud for premium features — without replacing what you already have.
Cloud ensemble models
Add OpenAI and Cohere embeddings alongside your local models for higher recall accuracy.
Cloud backup
Back up your local memories to OpenEngram Cloud. Peace of mind without giving up local control.
Cross-device sync
Access your memories from any device through the cloud dashboard. Your local instance stays the source of truth.
Add cloud when you need it
Start fully local. Link to OpenEngram Cloud later from Settings. No migration, no data loss.
Pick your path
Self-hosted gives you everything for free. Cloud adds managed convenience. Hybrid gives you both.
| Feature | Self-Hosted | Cloud | Hybrid |
|---|---|---|---|
| Local embedding models | ✅ All 4 models | — | ✅ All 4 models |
| Cloud ensemble models | — | ✅ OpenAI + Cohere | ✅ OpenAI + Cohere |
| Code Search | ✅ | — | ✅ |
| Dream Cycle | ✅ | ✅ | ✅ |
| Ensemble search | ✅ Local models | ✅ Cloud models | ✅ Both |
| Cloud backup | — | ✅ Automatic | ✅ |
| Cross-device sync | — | ✅ | ✅ |
| Setup required | One command | None | One command + link |
| Price | Free forever | Coming soon | Coming soon |
Your memories. Your control. Any model.
Talk to Claude on one device, ChatGPT on another, and Gemini on a third. Your Engram memories follow you everywhere.
Who is this for?
AI Coding Assistants
Your assistant remembers your codebase conventions, past decisions, and project context across sessions.
Customer Support Agents
Agents recall customer history, preferences, and past issues without re-reading every ticket.
Personal AI Companions
Your AI remembers your name, your family, your preferences — like a real relationship.
Multi-Agent Systems
Shared memory pools let agents collaborate with full context of what other agents have learned.
The ecosystem
Every piece is open source. Use them standalone or together.
Engram Core
Memory API with extraction, classification, scoring, and temporal reasoning
Engram Embed
Local embeddings on Metal GPU — multiple models, zero API costs
Engram Code
Semantic code search across your entire codebase
Dashboard
Real-time monitoring, health scores, knowledge graph visualization
MCP Server
Drop-in memory for Claude Desktop, Cursor, Windsurf, and any MCP client
TypeScript SDK
engram.remember() and engram.recall() — fully typed, 18 methods
Hosted cloud is coming soon
Cloud ensemble models, automatic backup, cross-device sync — all managed for you. Join the waitlist to get early access.
Self-hosting is available now with all features, completely free.