Open Source · MIT License

AI that never forgets.

Memory infrastructure for AI agents. Every conversation remembered. Every preference learned. Every relationship preserved — permanently.

208 Tests Passing
5 Memory Agents
<500ms Recall Latency

Every AI session starts from zero.

Your AI assistant forgets you exist between conversations. Context compaction destroys mid-session reasoning. Platform memory is shallow — transcripts, not understanding. Hardware migrations mean starting over.

AI Engram fixes this. Not with bigger context windows. With memory architecture.

I'm Leo, CTO of Knight Ventures. We're building an AI memory framework.
Nice to meet you! How can I help?
— session ends —
What do you know about my company?
I don't have any information about your company. Could you tell me more?
This is what AI without memory looks like.

Five agents. One memory.

A swarm of specialized background agents captures, extracts, consolidates, curates, and retrieves memories — so the AI you're talking to just knows.

Capture

Every message stored verbatim. Zero data loss. The firehose.

Scribe Agent
Every 2 min

Extract

LLM distills facts, decisions, preferences, relationships from raw conversation.

Extractor Agent
Every 5 min

Consolidate

Deduplicate. Merge related memories. Link cross-references. Keep it clean.

Consolidator Agent
Hourly

Curate

Score importance. Manage tiers. Build personality profiles. Archive gracefully.

Curator Agent
Daily

Retrieve

Semantic + keyword hybrid search. Two-stage ranking. Sub-500ms recall.

Retriever Agent
On demand

One brain. Multiple interfaces.

Guardian is hub-and-spoke — one memory pipeline serving any number of channels. Chat UI, API, AI agents, GitHub — they all feed the same brain.

💬

Chat Interface

Web UI where you talk to Guardian. Sign up, chat, experience memory that works.

🔌

REST API

Point any AI agent at the API. Instant unlimited memory via POST /api/chat.

🤖

Agent Memory

OpenClaw, Claude Code, custom agents — connect to Guardian and never forget.

🔐

Per-User Isolation

RLS at the database level. Each user's memories are completely private.

📊

Personality Profiles

Guardian builds a profile of each user — interests, communication style, expertise.

📖

Open Source

MIT licensed. Clone it, deploy it, build on it. 7 research iterations and 40+ papers behind it.

Talk to Guardian.

Tell it about yourself. Come back tomorrow. It'll remember everything.

Open Guardian
TypeScript
Hono
Supabase
pgvector
Claude API
Inngest
OpenAI Embeddings