Skip to content

tanerincode/synatyx

Repository files navigation

Synatyx

The memory layer your AI agents have been missing.

Give your LLM a persistent, structured, relevance-scored memory — that survives every conversation, every session, every project.

Python License: MIT MCP Docker


The Problem

LLMs are stateless. Every new conversation starts from zero — no memory of past decisions, preferences, or project context. You repeat yourself constantly. Your AI assistant forgets everything you taught it yesterday.

Synatyx fixes that.


What It Does

Synatyx is a Context Engine that plugs into any MCP-compatible AI client (Augment Code, Cursor, Claude Desktop, Claude Code) and gives it persistent, structured memory across all your conversations.

flowchart LR
    IDE(["🖥️ Your IDE\nAugment / Cursor / Claude"])
    MCP["⚙️ Synatyx"]
    LLM(["🤖 LLM"])

    IDE -->|MCP stdio| MCP
    MCP -->|relevant context injected| LLM

    subgraph Memory ["4-Layer Memory"]
        L1["🔴 Working Memory"]
        L2["🟠 Episodic Summaries"]
        L3["🟡 Semantic Knowledge"]
        L4["🟢 Permanent Rules"]
    end

    MCP <-->|read / write| Memory
Loading

Your AI now remembers what you decided last week, recalls how your codebase is structured, and follows your preferences without being told again.


Why Synatyx

🧠 Persistent memory across sessions Store facts, decisions, and context once — retrieve them forever. No more repeating yourself.

🎯 Relevance-ranked retrieval Hybrid dense + BM25 + MMR pipeline surfaces the right memories, not just the newest ones.

📦 Multi-project isolation Each project gets its own memory space. Switch projects, switch context — nothing bleeds over.

🔖 Checkpoints that never disappear Pin critical decisions as named snapshots. Deprecate when superseded — never permanently deleted.

✅ Persistent task tracking Tasks survive across sessions. Your AI picks up where it left off.

🤖 Agent skill registry Store, index, and RAG-search agent skill definitions. The right agent for the right task, automatically.

🏭 Production-ready Docker Compose, Alembic migrations, health checks, audit log — ready to deploy.


Works With

Client Integration
Augment Code MCP stdio
Cursor MCP stdio
Claude Desktop MCP stdio
Claude Code MCP stdio
Any MCP client JSON-RPC 2.0 / stdio

Get Started

git clone https://github.com/tanerincode/synatyx.git && cd synatyx
cp .env.example .env   # add your EMBEDDING_OPENAI_API_KEY
make                   # starts everything + tails logs

Full Setup Guide


Documentation

Doc What's inside
Local Setup Prerequisites, Docker, IDE config, Makefile reference, troubleshooting
MCP Tools Reference All 18 tools — params, descriptions, examples
Architecture 4-layer memory model, retrieval pipeline, tech stack, project structure

License

MIT © Taner Tombas

About

Give your LLM a persistent, structured, relevance-scored memory — that survives every conversation, every session, every project.

Resources

License

Stars

4 stars

Watchers

0 watching

Forks

Packages

 
 
 

Contributors

Languages