Persistent AI Memory

Every
agent
forgets.
Until now.

ContextMesh is a shared memory layer your AI agents can write to and read from — across tools, sessions, and your entire team. One API. Any agent.

Read the docs ↗

contextmesh — python sdk

# Any agent writes context

mesh.remember("prod DB is

postgres 15, us-east-1")

# Any other agent reads it

results = mesh.query(

"what do we know about

our database?"

)

# → [

{ score: 0.94,

text: "prod DB is postgres 15..." }

# ]

<50ms

Query latency p99

MCP

Native MCP server included

∞

Cross-agent, cross-tool

The Problem

AI agents are powerful.
But they're amnesiac.

Cursor doesn't know what Claude decided last week. Claude doesn't know what your Cursor session learned yesterday. Every new conversation, every new tool, every new session — they all start from zero.

You repeat yourself constantly. "As I mentioned before, our API uses camelCase." "Like I said, the DB is on AWS." You're the shared memory layer. You shouldn't have to be.

ContextMesh is the RAM layer above all your agents.
Write once. Any agent reads it. Always.

Four endpoints.
That's the whole API.

No learning curve.
10 min integration.

◈

POST /remember

An agent writes a piece of context. It's embedded into a 1536-dim vector and stored with your metadata — tags, agent name, timestamp, confidence score.

⌕

POST /query

Any agent asks a natural language question. ContextMesh does semantic similarity search and returns the top-N most relevant entries, ranked by score.

DELETE /forget

When context becomes stale, remove it by ID. Clean up outdated decisions or wrong assumptions before they confuse your agents.

≡

GET /list

Browse your entire context store. Filter by tag. Audit what your agents know. Paginate through the full history with limit + offset.

Works where
you already work.

Python SDK, JavaScript SDK, MCP server, or raw REST. Integrates with Cursor, Claude Code, Cline, and anything else that speaks MCP.

◈

MCP Server included

One line in mcp.json and Claude/Cursor gain persistent memory automatically.

◈

Namespaced workspaces

Each workspace is isolated. Team plans get 10 workspaces. Enterprise is unlimited.

◈

Semantic search, not keyword

"What do we know about our DB?" finds "postgres 15 on AWS" even without keyword overlap.

◈

Tag, filter, expire

Tag entries by type. Filter queries to specific domains. Keep context clean and relevant.

Python

JavaScript

MCP

cURL

# pip install contextmesh
from contextmesh import Mesh

mesh = Mesh("cm_live_your_key")

# Store context
mesh.remember(
  "API rate limit is 1000 req/min",
  tags=["infra"],
)

# Query it from any agent
hits = mesh.query("rate limits?")
# → [{ text: "API rate limit...", score: 0.96 }]
    

// npm install contextmesh
import { Mesh } from "contextmesh";

const mesh = new Mesh("cm_live_your_key");

// Store context
await mesh.remember("Use camelCase for JS", {
  tags: ["conventions"],
});

// Query it later
const hits = await mesh.query("code style?");
    

// ~/.cursor/mcp.json
{
  "contextmesh": {
    "command": "npx",
    "args": ["contextmesh-mcp"],
    "env": {
      "CM_KEY": "cm_live_your_key"
    }
  }
}

// Claude/Cursor now has tools:
//  remember(text, tags?)
//  recall(query, top_k?)
//  forget(id)
//  list_context()
    

# Store context
curl -X POST https://contextmesh.dev/remember \
  -H "Authorization: Bearer cm_live_..." \
  -d '{"text":"prod DB is postgres 15",
     "tags":["database"]}'

# Query
curl -X POST https://contextmesh.dev/query \
  -H "Authorization: Bearer cm_live_..." \
  -d '{"q":"database info?","top_k":3}'
    

Getting Started

Up in 10 minutes.
Works from anywhere.

Four steps from zero to persistent memory across all your AI agents.

Get your API key

Pick a plan below → pay via Paddle → key arrives in your email instantly.

cm_live_U7f9x...

Store your first context

One POST request. Text gets embedded and stored with your metadata.

curl -X POST https://contextmesh.dev/remember \
  -H "Authorization: Bearer cm_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{"text":"Auth uses RS256 JWT, 15min expiry",
       "tags":["auth","security"]}'

# → {"id": "abc-123", "status": "stored"}

Query from any agent

Ask in plain English. Semantically matched — no keyword overlap needed.

curl -X POST https://contextmesh.dev/query \
  -H "Authorization: Bearer cm_live_your_key" \
  -H "Content-Type: application/json" \
  -d '{"q":"how does login work?","top_k":5}'

# → [{text: "Auth uses RS256 JWT...", score: 0.94}]

Add MCP — zero-effort memory

One block in your MCP config. Cursor and Claude get remember, recall, and forget as native tools.

// ~/.cursor/mcp.json  or  ~/.claude/mcp.json
{
  "contextmesh": {
    "command": "npx",
    "args": ["contextmesh-mcp"],
    "env": { "CM_KEY": "cm_live_your_key" }
  }
}

Simple pricing.

No per-token fees. No per-request surprises. Flat monthly rate by plan.

Solo

$19

/ month

◈ 1 workspace

◈ 500k context entries

◈ 100k queries / month

◈ Python + JS SDK

◈ MCP server

Get started →

Popular

Team

$99

/ month

◈ 10 workspaces

◈ 5M context entries

◈ 1M queries / month

◈ Everything in Solo

◈ Usage analytics

Get started →

Enterprise

$599

/ month

◈ Unlimited workspaces

◈ Unlimited entries

◈ Unlimited queries

◈ On-prem option

◈ SLA + dedicated support

Everyagentforgets.Until now.

AI agents are powerful.But they're amnesiac.

Four endpoints.That's the whole API.

Works whereyou already work.

Up in 10 minutes.Works from anywhere.

Simple pricing.

Start remembering.

Every
agent
forgets.
Until now.

AI agents are powerful.
But they're amnesiac.

Four endpoints.
That's the whole API.

Works where
you already work.

Up in 10 minutes.
Works from anywhere.