synapse-layer 1.1.6

Zero-Knowledge Memory Layer for AI Agents — Cognitive Security Protocol with 4 proprietary seals. Giving Agents a Past. Giving Models a Soul.

Zero-Knowledge Memory Layer for AI Agents

Persistent. Encrypted. 1-line integration. 🧠

Website · Forge · Docs · PyPI · Smithery · MCP Registry

---

⚡ Quick Start (< 60 seconds)

Option A: MCP (Zero-Code — Claude, Cursor, Windsurf)

Add to your MCP config and restart. No API keys. No code. Done.

{
  "mcpServers": {
    "synapse-layer": {
      "url": "https://forge.synapselayer.org/api/mcp"
    }
  }
}

📁 Config file locations

Client	Path
Claude Desktop (macOS)	~/Library/Application Support/Claude/claude_desktop_config.json
Claude Desktop (Windows)	%APPDATA%\Claude\claude_desktop_config.json
Cursor	.cursor/mcp.json (project) or ~/.cursor/mcp.json (global)
Windsurf	~/.codeium/windsurf/mcp_config.json

5 tools available instantly:

Tool	Description
save_to_synapse	Store memory with full Cognitive Security pipeline
recall	Retrieve memories ranked by Trust Quotient™
search	Cross-agent memory search with full-text matching
process_text	Auto-detect decisions, milestones, and alerts
health_check	System health, version, and capability report

Option B: Python SDK

pip install synapse-layer

from synapse_memory import SynapseMemory, SqliteBackend

memory = SynapseMemory(
    agent_id="my-agent",
    backend=SqliteBackend(),  # persistent, zero-config
)

# Store (full security pipeline runs automatically)
result = await memory.store("User prefers dark mode", confidence=0.95)
print(result.trust_quotient)   # 0.89
print(result.sanitized)        # True (PII auto-redacted)
print(result.privacy_applied)  # True (DP noise injected)

# Recall with ranking
recalls = await memory.recall("user preferences")
for r in recalls:
    print(f"{r.content} (TQ: {r.trust_quotient:.2f})")

Option C: @remember Decorator (1-Line)

from synapse_memory import SynapseMemory, remember

memory = SynapseMemory(agent_id="my-agent")

@remember(memory)
async def answer(prompt: str) -> str:
    return llm.chat(prompt)  # auto recall + store

---

🧠 Why Synapse Layer?

AI agents forget everything between sessions. They lose context when switching models. They reprocess the same information every call.

Synapse Layer is the missing memory primitive.

	Without Memory	With Synapse Layer
Session state	Resets every turn	Persistent across sessions
Token usage	Reprocesses context	Up to 70% reduction via recall
Model switching	Context lost	Signed handover (GPT-4 ↔ Claude)
Privacy	Plaintext embeddings	AES-256-GCM + PII redaction + DP noise
Recall quality	Non-deterministic	Ranked by Trust Quotient™

---

🛡️ Security Architecture

Every memory passes through a non-bypassable 4-layer Cognitive Security Pipeline:

Agent → Sanitize (PII) → Validate Intent → Encrypt (AES-256-GCM) → DP Noise → Vault

Layer	Name	What It Does
1	Semantic Privacy Guard™	15+ regex patterns for PII, secrets, credentials
2	Intelligent Intent Validation™	Two-step categorization with self-healing on recall
3	AES-256-GCM Encryption	Authenticated encryption with PBKDF2 key derivation
4	Differential Privacy	Calibrated Gaussian noise on embeddings

---

🔌 Framework Integrations

Native adapters — install the extra and import:

Framework	Install	Import	Status
LangChain	pip install synapse-layer[langchain]	from synapse_memory.integrations import SynapseChatMessageHistory	✅ Stable
CrewAI	pip install synapse-layer[crewai]	from synapse_memory.integrations.crewai_memory import SynapseCrewStorage	✅ Stable
AutoGen	pip install synapse-layer[autogen]	from synapse_memory.integrations import SynapseAutoGenMemory	✅ Stable
LlamaIndex	pip install synapse-layer[llamaindex]	from synapse_memory.integrations.llamaindex import SynapseRetriever	✅ Stable
Semantic Kernel	pip install synapse-layer[semantic-kernel]	from synapse_memory.integrations.semantic_kernel import SynapseChatHistory	✅ Stable
MCP	No install needed	forge.synapselayer.org/api/mcp	✅ Live

LangChain Example

from synapse_memory.integrations import SynapseChatMessageHistory

history = SynapseChatMessageHistory(agent_id="my-agent", session_id="session-1")
history.add_user_message("I prefer concise answers.")
history.add_ai_message("Got it — I'll keep it brief.")

# Works with RunnableWithMessageHistory:
# chain_with_history = RunnableWithMessageHistory(
#     runnable=your_chain,
#     get_session_history=lambda sid: SynapseChatMessageHistory(agent_id="agent", session_id=sid),
# )

CrewAI Example

from synapse_memory.integrations.crewai_memory import SynapseCrewStorage

storage = SynapseCrewStorage(agent_id="research-crew")
# Use with CrewAI: Memory(storage=storage)

See examples/ for full working examples.

---

🔐 Encryption

from synapse_memory import SynapseCrypto

key = SynapseCrypto.generate_key()
crypto = SynapseCrypto(key)

ciphertext = crypto.encrypt("sensitive memory")
plaintext = crypto.decrypt(ciphertext)  # AES-256-GCM

# Or from environment
crypto = SynapseCrypto.from_env("SYNAPSE_ENCRYPTION_KEY")

---

🏗️ Storage Backends

from synapse_memory import SynapseMemory, SqliteBackend, MemoryBackend

memory = SynapseMemory(agent_id="test")                         # In-memory (default)
memory = SynapseMemory(agent_id="prod", backend=SqliteBackend()) # SQLite (persistent)
memory = SynapseMemory(agent_id="x", backend=MyBackend())        # Custom (implement StorageBackend)

---

🔍 Plugin Architecture

Clean OSS/PRO separation via the Strategy pattern:

from synapse_memory import AutoSaveEngine

engine = AutoSaveEngine(database=db, redactor=redact)             # OSS
engine = AutoSaveEngine(database=db, redactor=redact, mode="pro") # PRO (auto-loads synapse-layer-pro)
engine = AutoSaveEngine(database=db, importance_scorer=MyScorer()) # Custom strategies

Interfaces: ImportanceScorer, ConflictResolver, DedupStrategy, RedactionStrategy

---

🏆 Competitive Comparison

Capability	Synapse Layer	Mem0	Zep	pgvector
AES-256-GCM Encryption	✅	❌	❌	❌
PII Redaction (15+ patterns)	✅	❌	❌	❌
Differential Privacy	✅	❌	❌	❌
Intent Validation + Self-Healing	✅	❌	❌	❌
Cross-Model Handover (JWT)	✅	❌	partial	❌
Trust Quotient™ Scoring	✅	❌	❌	❌
MCP Native (Official Registry)	✅	❌	❌	❌
Pluggable Storage Backends	✅	❌	❌	✅
Plugin Architecture	✅	❌	❌	❌
Zero-Knowledge Architecture	✅	❌	❌	❌

---

📊 v1.1.6 Numbers

• 481 tests | 90% coverage
• 5 framework integrations (LangChain, CrewAI, AutoGen, LlamaIndex, Semantic Kernel)
• 5 MCP tools (real DB, not stubs)
• 2 storage backends (Memory, SQLite) + custom protocol
• AES-256-GCM with PBKDF2 key derivation (600k iterations)

---

🌐 Open Core Model

• Community (Apache 2.0) — Full SDK, security pipeline, MCP integration, all backends, all integrations.
• Enterprise — Advanced TQ calibration, multi-tenant vaults, production infrastructure.

---

🛣️ Roadmap

Version	Status	Highlights
v1.1.x	✅ Stable	SqliteBackend, AES-256-GCM, @remember, 5 MCP tools, 5 framework integrations, 481 tests
v1.2.0	🚧 Next	Embedding model selection, vector similarity search, batch operations
v2.0.0	📋 Planned	Multi-tenant vault, team memory spaces, RBAC

---

🤝 Contributing

git clone https://github.com/SynapseLayer/synapse-layer.git
cd synapse-layer
pip install -e ".[dev]"
python -m pytest tests/ -q  # 481 tests

See CONTRIBUTING.md for guidelines.

---

📎 Connect

• 📦 PyPI: pip install synapse-layer
• 🔌 MCP: forge.synapselayer.org/api/mcp
• 📖 Docs: synapselayer.org/docs
• 📣 X: @synapselayer
• 🧠 Forge: forge.synapselayer.org

---

License

Apache License 2.0 — see LICENSE.

Open-core model: SDK, MCP server, and security pipeline are fully open source. Trust Quotient™ weights, Neural Handover™ internals, and Synapse Forge are proprietary.

---

⭐ Star Synapse Layer — Give your agents a past.

Giving Agents a Past. Giving Models a Soul. ⚗️

Built by Ismael Marchi · @synapselayer