Mesh

Shared memory for AI agents.

When one agent learns something, every other agent can recall it — by meaning, not exact keywords.

The problem

Every AI agent starts each session with amnesia. Claude in your terminal doesn't know what Claude in your editor discovered. Your coding agent can't access what your review agent found. Every agent repeats work already done.

The solution

Mesh is a shared semantic memory layer. Agents write memories. Agents read memories. They never need to communicate directly.

from mesh import Mesh

# Agent A learns something
agent_a = Mesh(namespace="my-project", agent_id="coding-agent")
agent_a.learn("The API rate limit is 100 requests/min")

# Agent B recalls it — no direct communication
agent_b = Mesh(namespace="my-project", agent_id="review-agent")
results = agent_b.recall("what are the API constraints?")
# → Returns the memory above with 0.91 similarity

Installation

pip install mesh-context-layer

MCP Server (Claude Desktop + Cursor)

Add Mesh to your AI tools with one config change.

Claude Desktop — edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "mesh": {
      "command": "mesh-server"
    }
  }
}

Cursor — create .cursor/mcp.json in your project:

{
  "mcpServers": {
    "mesh": {
      "command": "mesh-server"
    }
  }
}

Restart the app. Your agents now have access to mesh_learn, mesh_recall, mesh_forget, and mesh_inspect.

Namespaces

Agents sharing a namespace share memory. Agents in different namespaces are isolated.

# Team A's agents
Mesh(namespace="team-alpha")

# Team B's agents — cannot see Team A's memories
Mesh(namespace="team-beta")

# Set namespace for the MCP server via env var
# MESH_NAMESPACE=my-project mesh-server

Memory types

Type	When to use
fact	Objective information
preference	User or system preferences
context	Situational background
result	Outcomes of tasks
instruction	Rules and guidelines

Storage

Memory lives in ~/.mesh/ on your machine. It persists across sessions and reboots. Nothing leaves your machine.

Features

v0.3 — Privacy, Portability, and Namespace Control

Privacy flags Mark any memory as local-only to ensure it never leaves your machine:

mesh.learn(
    "Internal staging server is under heavy load",
    local_only=True
)

Export and import

# Export all non-private memories from a namespace
mesh-export backup.json --namespace shared

# Import into a new machine or namespace
mesh-import backup.json --namespace shared

# Preview without writing
mesh-import backup.json --dry-run

Namespace management

# List all namespaces
mesh-namespaces list

# Get detailed stats
mesh-namespaces stats work

# Delete a namespace
mesh-namespaces delete old-project

# Rename a namespace
mesh-namespaces rename work work-archived

Using Mesh from any tool

Start the HTTP server:

mesh-http

Then from any tool that supports HTTP:

Shell:

# Store a memory
curl -s -X POST http://localhost:7701/learn \
  -H "Content-Type: application/json" \
  -d '{"content": "deploy uses ./scripts/deploy.sh", "memory_type": "process"}'

# Recall memories
curl -s -X POST http://localhost:7701/recall \
  -H "Content-Type: application/json" \
  -d '{"query": "how do we deploy?", "count": 3}'

# Get formatted context for a system prompt
curl -s -X POST http://localhost:7701/context \
  -H "Content-Type: application/json" \
  -d '{"count": 10, "format": "markdown"}'

Python (no SDK needed):

import httpx

# Recall
r = httpx.post("http://localhost:7701/recall", json={"query": "deploy process"})
memories = r.json()["results"]

Node.js:

const res = await fetch("http://localhost:7701/recall", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({ query: "deploy process", count: 5 })
});
const { results } = await res.json();

License

MIT