Trinity Memory — A Triune Architecture for AGI Long-Term Memory

A high-performance, production-ready persistent memory layer for AI agents. Trinity integrates 12+ state-of-the-art memory approaches into a unified architecture with 50-tier guardian chains, 47 retrieval channels, and multi-modal support.

中文版 README → README.zh.md

Quick Start

pip install trinity-memory

from trinity import Trinity

mem = Trinity()
mem.ingest("User prefers dark mode", tags=["preference", "ui"])
results = mem.search("user preference")
print(results)

CLI

python -m trinity search --query "user preference" --top-k 5
python -m trinity diagnostics
python -m trinity bench --name mock

MCP Server

{
  "mcpServers": {
    "trinity-memory": {
      "command": "trinity-mcp",
      "args": ["--mode", "stdio"]
    }
  }
}

Related MCP server: Strata Memory MCP Server

Architecture

Trinity is built on three core layers, integrating cutting-edge memory research:

Benchmarks

Features

• Multi-Modal: Text, image, and audio memory in a unified interface

• Multi-Tenant: Three-level isolation (persona_id / session_id / tenant_id)

• 47 Retrieval Channels: Progressive cascading from 0.05ms P50

• 50-Level Guardian Chain: L1-L50 with reasoning drift detection

• MCP Support: Standard Model Context Protocol (stdio + SSE)

• REST API: FastAPI with 8 endpoints + Web Dashboard

• Multiple Backends: SQLite, PostgreSQL, ChromaDB, Vectile

• Self-Evolution: Auto-curricula, Engram memory, Consolidation sleep

• Knowledge Graph: Semantic / Relational / Temporal graph queries

• Docker Ready: docker compose up -d for one-click deployment

Deployment

Docker

docker build -t trinity-memory .
docker run -d -p 8100:8100 -p 8000:8000 -v /data:/data trinity-memory

Docker Compose

docker compose up -d

REST API

# Write memory
curl -X POST http://localhost:8100/memories \
  -H "Content-Type: application/json" \
  -d '{"content":"User info","importance":0.8}'

# Search memory
curl "http://localhost:8100/search?q=user&top_k=5"

Commercial

Documentation

Full documentation: https://trinity-tick.github.io/trinity

• Getting Started

• Architecture

• API Reference

• Benchmarks

• Deployment

License

MIT License — free for commercial and non-commercial use.

Star History