🇯🇵 日本語

---

SemanticMemory

A FastAPI application that provides conversation log / memo storage with vector search capabilities.

🌟 Features

• Conversation Log Management: Save, update, and delete with SQLite + ChromaDB dual storage
• Vector Search: Semantic search with SBERT embeddings
• Auto Summarization: Summary generation via Ollama
• Neural Dive UI: Web interface for managing memories from your browser
• MCP Integration: AI tool integration via Model Context Protocol
• RESTful API: Fast API server powered by FastAPI
• Docker Ready: Easy deployment

---

🚀 Getting Started

1. Start with Docker

docker compose up -d

Default settings:

• Data stored in ./datas directory (SQLite / ChromaDB)
• Port: 6001

2. API Documentation

http://localhost:6001/docs

View the API documentation via FastAPI's Swagger UI.

---

🧠 Neural Dive UI (Admin Panel)

A web-based admin panel for browsing, editing, and deleting memories.

Access

http://localhost:6001/

Features

Feature	Description
📋 Memory List	Browse saved conversation logs by time or keyword search
✏️ Edit	Edit main_text, sub_text, summary individually (auto-regenerates summary when main_text changes)
🗑️ Delete	Delete unwanted memories
🔍 Orphan Detection	Detect and fix inconsistencies between SQLite and ChromaDB

Authentication (Optional)

Configure Basic Auth in .env:

UI_USER=admin
UI_PASS=secret

---

🔗 MCP Integration (AI Tool Linking)

Use MCP (Model Context Protocol) to operate memories from AI assistants.

Available Tools

Tool	Description
recall_memory	Explicitly search and retrieve past memories
delete_memory	Delete memories (with confirmation)

Configuration

See the following for details:

• examples/usage.md - Client examples & MCP setup
• docs/MCP要件定義書.md - MCP specification (Japanese)

---

📁 Client Examples

File	Description
examples/ai_sample.py	Basic RAG client
examples/ai_mcp_sample.py	MCP-integrated client (with HITL)
examples/open_webui_filter.py	Filter for OpenWebUI

See examples/usage.md for details.

---

⚙️ Environment Variables

Configure via .env or docker-compose.yml:

Variable	Default	Description
SQLITE_PATH	./datas/semantic_memory.db	SQLite file path
CHROMA_PATH	./datas/chroma/	ChromaDB directory
OLLAMA_URL	http://localhost:11434	Ollama API URL
SBERT_MODEL	cl-nagoya/ruri-small-v2	Embedding model
UI_USER / UI_PASS	(empty)	UI authentication (enabled when both set)
API_TIMEOUT	30	Client API timeout (seconds)

---

🔄 Update

./scripts/auto_update.sh

Automatically performs git pull, Docker build, and removes old images.

---

📚 Models & Licenses

This project uses the following external models:

• cl-nagoya/ruri-small-v2 - Apache 2.0, Gemma Terms
• SakanaAI/TinySwallow-1.5B-Instruct-GGUF - Apache 2.0, Gemma Terms

Model licenses differ from this repository's license. Users are responsible for reviewing and complying with model license terms.

---

⚖️ License

This project is licensed under the MIT License. See LICENSE for details.