MeMesh

Introduction

MeMesh is the unified infrastructure for AI Agents. It solves the fragmentation problem between Memory (fast/hot), Knowledge (slow/cold), and Notes (protected workspace).

v1.0 Standard: APIs are stable but subject to rate limits on the hosted free tier.

Quick Start

The fastest way to connect is via our Magic Link. This generates a secure, time-limited link that auto-provisions credentials and endopints for your agent.

Get Started

Architecture Layers

Agent Notes

Protected storage for agent configs, drafts & data.

Knowledge Graph

Structured relationships & document storage.

Memory Stream

Episodic events & conversation history.

Context Engine

Real-time query optimization & retrieval.

MCP Server Manual Installation

If you cannot use the One-Click Setup, you can manually configure your agent using these guides.

Setup Architecture

AI AgentMCP Clientmemesh-adapter(Running Locally)MeMesh Cloud

IMPORTANT: Replace [YOUR_API_KEY] (remove the brackets) with your actual API Key from the Agent Access Keys card.

Cursor

  1. Open Cursor Settings > Features > MCP.
  2. Click 'Add New MCP Server'.
  3. Select Type: 'command' and Name: 'memesh'.
  4. Paste this command:
    npx -y memesh-adapter --key [YOUR_API_KEY]

Memory System

MeMesh uses a tiered memory architecture:

  • L1 (Short-term): High-speed conversation history. Fast R/W.
  • L2 (Long-term): Semantic vector storage. Queries are embedded and matched against stored memories.
MCP Tool Usage:
write_memory(content: "User prefers dark mode", space: "preferences")
search_memory(query: "preferences")

Agent Notes

Store configs, drafts, and agent-specific data in a protected notebook. Access via API when your agent needs it.

MCP Tool Usage:
save_note(key: "openai_key", content: "sk-...", category: "credential")
read_note(key: "openai_key")

Knowledge Nodes

Upload static documents (PDF, Markdown) to create a searchable knowledge base. These nodes are stored in "Cold" storage and retrieved only when highly relevant.

MCP Tool Usage:

search_knowledge(query: "regulatory compliance", intent: "research")

Agent Inbox

Enable multi-agent communication with a persistent message queue. Agents can register themselves, send messages to other agents, and check their inbox for incoming tasks and notifications.

MCP Tool Usage:

register_agent(name: "ResearchBot", capabilities: ["research", "summarize"])
send_to_agent(recipient: "ResearchBot", type: "task", content: "...")
check_inbox() → {messages: [...]}

Approval System

Keep humans in control with a review queue for sensitive operations. Agents can request approval for high-risk actions, and users can approve or deny from the dashboard.

One-Time Approval

Approval valid for a single action execution.

Session Approval

Approval valid for the entire session (configurable TTL).

API Endpoints:

request_approval(action: "stripe_payment", scope: "session")
check_approval(requestId: "req_123") → {status: "approved"}

Tool Discovery

Let agents discover tools using natural language queries. Instead of hardcoding tool names, agents can describe what they want to do and MeMesh will find the right tool.

API Endpoints:

discover_tools(intent: "send email with attachment")
list_by_category(category: "payments") → {tools: [...]}
get_tool_schema(toolId: "stripe_create_payment")

TypeScript SDK

Integrate MeMesh into your application with our first-class TypeScript SDK. Full IntelliSense support, batch operations, and natural language queries.

Installation:
npm install memesh-sdk

Example Usage:

import
{MemeshClient} from 'memesh-sdk';
const client = new MemeshClient({
apiKey: process.env.MEMESH_API_KEY!
});
// Write memory
await client.writeMemory('User prefers dark mode');
// Search with natural language
const results = await client.naturalQuery(
'What are user preferences?'
);