synapse-layer

name	synapse-layer
description	Persistent memory infrastructure for AI agents
version	2.3.7
author	Ismael Marchi
tags	["memory","mcp","ai-agents","llm","typescript","cross-agent"]

Synapse Layer — Memory Skill

What This Does

Synapse Layer is a persistent memory infrastructure for AI agents. It solves the stateless-agent problem: agents forget decisions, preferences, and context across sessions. With Synapse, any MCP-compatible agent can store and recall memories — AES-256-GCM encrypted at rest, scored by Trust Quotient for reliability, and scoped per-user with tenant isolation. One agent stores, another recalls. Memory survives across sessions, models, and tools.

Agent Rules

MUST:

• store() after each interaction that contains a decision, preference, or constraint worth persisting.
• recall() before responding when the user references prior context, preferences, or past work.
• Filter recall results by tq >= 0.7 when high confidence is required.
• Include agent identifier on every store() call for cross-agent traceability.

MUST NOT:

• Store PII without explicit user consent.
• Assume a memory exists without calling recall() first.
• Expose raw memory content in logs or error messages.
• Store ephemeral or throwaway content (e.g., "ok", "thanks").

Trust Quotient (TQ)

Trust Quotient is a per-memory confidence score ranging from 0.0 to 1.0.

• What it measures: Reliability of a memory based on content density, semantic alignment with recall queries, and noise level.
• How to use: Filter recall results by tq >= 0.7 for high-confidence decisions. Use tq >= 0.5 for general context. Memories below 0.3 may contain noise or low-relevance content.
• Scoring: TQ = f(density, alignment, noise). Weights are proprietary and dynamically calibrated.

Multi-LLM Compatibility

Synapse Layer works with any MCP-compatible client:

• Claude (Desktop + API) — native MCP support
• GPT-4 / GPT-4o — via Synapse Proxy (OpenAI SDK compatible)
• Gemini — via MCP bridge
• Llama / Mistral / local models — via MCP stdio bridge
• LangChain / CrewAI / AutoGen — via Python SDK adapters

Links

• Forge: forge.synapselayer.org
• Docs: forge.synapselayer.org/docs
• SDK (Python): pypi.org/project/synapse-layer
• GitHub: github.com/SynapseLayer