# Synapse Layer — Forge

> Persistent memory infrastructure for AI agents — encrypted, governed, and cross-agent.
> The OAuth for AI Memory. MCP-native Trust Infrastructure for AI Agents.
> Version: 2.4.4
> Audience: AI agents and applications that USE the Synapse memory service at runtime.

Forge is the hosted control plane and MCP endpoint for Synapse Layer. If you are an AI
agent or client (Claude, GPT, Gemini, Cursor, Perplexity, Windsurf, Cline, Continue.dev,
or any MCP-compatible runtime), this file tells you how to connect, authenticate, and
operate against persistent memory.

## What Synapse Layer is
Persistent, encrypted, governed memory for AI agents that survives across sessions,
models, and tools. Every memory is AES-256-GCM encrypted at rest with a per-operation
random IV and a 128-bit GCM authentication tag validated on every read, scored by
Trust Quotient (0.0-1.0), and isolated per tenant (1 user = 1 tenant = 1 private mind).

## How to connect (MCP)
- [MCP endpoint](https://forge.synapselayer.org/api/mcp): JSON-RPC over streamable-http for MCP-compatible clients
- [Health endpoint](https://forge.synapselayer.org/api/health): public HTTP GET proof-of-life returning status and version (no auth)
- Transport: streamable-http
- Auth: Bearer token (Connect Token, scheme `sk_connect_`)
- Public method (no auth): initialize
- Authenticated methods: tools/list, tools/call
- [Server card](https://forge.synapselayer.org/.well-known/mcp/server-card.json): MCP server capability descriptor
- [Agent card](https://forge.synapselayer.org/.well-known/agent.json): Agent discovery metadata
- [Get a Connect Token](https://forge.synapselayer.org/dashboard/tokens): Issue a Bearer token (sk_connect_) from the dashboard

## MCP tools (13)
- recall - retrieve relevant persisted memory (semantic, temporal, priority, hybrid routing)
- save_to_synapse - persist memory with encryption at rest, sanitization, and dedup controls
- process_text - extract candidate memories from free-form text with governance filters
- search - search persisted memory across agent scopes using full-text matching
- health_check - check service availability, engine version, and storage health (public)
- initialize_context - initialize a persistent memory context for a session or agent
- save_memory - save a memory entry (alias of save_to_synapse)
- store_memory - store structured memory with metadata and trust scoring
- recall_memory - recall persisted memory by query (alias of recall)
- list_memories - list memory metadata with pagination and governance limits
- memory_feedback - submit feedback to adjust trust scoring for stored memories
- neural_handover - transfer contextual state between agents with continuity controls
- slo_report - return uptime and SLO metrics for the MCP service

## SDK (alternative to raw MCP)
```
pip install synapse-layer
```
```python
from synapse_layer import Synapse
client = Synapse(api_key="sk_connect_...")
client.remember("User prefers dark mode")
memories = client.recall(query="user theme preference", top_k=5)
```

## Security & governance
- Encryption: AES-256-GCM at rest, per-operation random IV, 128-bit GCM tag validated on every read
- Auth: OAuth 2.0 + PKCE S256; userId always from auth context, never from request body
- Tokens: SHA-256 hashed; raw tokens never persisted in the database
- Isolation: per-tenant; zero cross-user memory access
- Privacy: PII sanitization before storage; designed for LGPD/GDPR alignment (not formally certified)

## Pricing
- Free: 100 memories, 1 agent, $0/month
- Pro: 10,000 memories, unlimited agents, $29/month
- Enterprise: custom limits and dedicated infrastructure, contact sales

## Canonical ecosystem (distributed knowledge graph)
- [Website](https://synapselayer.org): Product overview and positioning
- [Forge (control plane)](https://forge.synapselayer.org): Hosted MCP endpoint and dashboard
- [Docs](https://forge.synapselayer.org/docs): Integration guides and API reference
- [Python SDK (PyPI)](https://pypi.org/project/synapse-layer/): pip install synapse-layer
- [TypeScript SDK (npm)](https://www.npmjs.com/package/synapse-layer): npm install synapse-layer
- [Source (core)](https://github.com/SynapseLayer/synapse-layer): Core engine and MCP server source
- [Python SDK repo](https://github.com/SynapseLayer/synapse-sdk-python): Python client source
- [LangGraph integration](https://github.com/SynapseLayer/synapse-layer-langgraph): LangGraph adapter for Synapse memory
- [Agent skill](https://github.com/SynapseLayer/synapse-layer-skill): Reusable agent skill package
- [Full technical spec](https://github.com/SynapseLayer/synapse-layer/blob/main/llms-full.txt): Extended llms-full.txt specification