🧠
Engram
Dual-Memory System · Episodic + Semantic · LLM Reasoning · Agent Memory
🔌
Interfaces
⌨️
CLI (Typer)
engram remember / recall / think
🔧
MCP Server
10 tools for AI agents
🌐
HTTP API
FastAPI REST endpoints
⚡
WebSocket
Real-time streaming
⚙️
Core Engine
🧠
Memory Manager
remember() / recall() / think() / ingest()
💭
Reasoning Engine
Combine episodic + semantic for thinking
🔍
Hybrid Search
Vector similarity + FTS + Graph traversal
👁️
Watcher Daemon
Auto-capture OpenClaw & Claude Code sessions
🔗
Entity Extractor
LLM-powered entity & relation extraction
💾
Data Layer
📊
Qdrant
Vector embeddings · Episodic memory
🕸️
NetworkX
Semantic graph · Entities & relations
🔎
SQLite FTS5
Full-text search · Keyword matching
⚡
Redis
Cache · Pub/Sub (optional)
📍 Data Flow
REMEMBER
Input
→
Embed
→
Store in Qdrant
→
Extract entities
→
Update Graph
RECALL
Query
→
Vector search
→
FTS search
→
Merge & Rank
→
Return
THINK
Question
→
Recall relevant
→
Query graph
→
LLM reasoning
→
Answer
INGEST
Text
→
LLM extract
→
Add to graph
→
Store memory
🤖
LiteLLM
Claude
GPT-4
Gemini
Mistral
🦙
Ollama
gemma3
qwen3.5
llama3
phi-4
📐
Embeddings
text-embedding-3
nomic-embed
bge-m3
10
MCP Tools
4
Interfaces
3
Memory Types
∞
Memories