Synapse Layer

Zero-Knowledge Memory Layer for AI Agents

Giving Agents a Past. Giving Models a Soul.

Website • Docs • GitHub

---

The Problem

AI agents are stateless by design. They forget everything between sessions or when switching models. This leads to repeated questions, token waste, and poor user experience.

The Solution

Synapse Layer is a persistent, encrypted, cross-model memory infrastructure with cryptographic guarantees.

Core Capabilities

• 🔐 Zero-Knowledge — AES-256-GCM client-side encryption (you own the keys)
• 🤝 Neural Handover™ — Secure context transfer between Claude, GPT, Llama, Gemini, etc.
• ⚖️ Trust Quotient™ — Automatic weighted conflict resolution
• 🔌 MCP Native — Works natively with Claude Desktop, LangChain, CrewAI
• 📦 Open SDK — Apache 2.0. No vendor lock-in.

Now available as MCP Skill → inference-sh/skills PR #8

---

Quick Start

pip install synapse-layer

import asyncio
from synapse_layer import SynapseMemory

async def main():
    memory = SynapseMemory(agent_id="my-agent")

    await memory.store(
        content="User prefers concise technical answers in Brazilian Portuguese",
        confidence=0.97
    )

    results = await memory.recall("user language preference")
    
    for item in results:
        print(f"→ {item['content']} (TQ: {item['trust_quotient']:.3f})")

asyncio.run(main())

---

How It Works

Your Agent → Synapse SDK (encrypts locally) → Encrypted Memory Vault → Never sees plaintext.

---

Security First

See SECURITY.md for full architecture, key management and compliance (GDPR / LGPD ready).

---

License

Apache License 2.0 — see LICENSE.

---

Giving Agents a Past. Giving Models a Soul. ⚗️