Leave an imprint on your IDEs.

One persistent memory layer for Claude Code, Cursor, Codex, Antigravity — and every MCP-capable IDE. Automatic capture. Semantic recall. Real-time contradiction detection. Your full context, everywhere you build.

Vercel AI SDK IntegrationAWS Serverless InfrastructureSub-100ms Hydration

Quick start

Up and running
in under 2 minutes.

Get started free
0130 sec

Sign in

Create your free Imprint account. No credit card, no install yet.

0260 sec

Connect your IDE

Paste one config block into Claude Code, Cursor, or Codex. MCP activates instantly.

03Forever

AI remembers you

Switch IDEs, start a new chat — your full context is always there, automatically.

Claude Code · Cursor · Codex · Antigravity · VS Code · Windsurf

Powered by Cognee Cloud

Your memory isn't a list —
it's a graph.

Every fact you teach Imprint is remembered into Cognee Cloud, which connects it to everything related — so recall reasons over the whole graph, not one row at a time. Drag-free, live, and interactive:

Cognee Knowledge Graph powered by Cognee

Cognee structures your memories into a graph, not isolated rows — nodes cluster by topic, teal links are shared-context connections, red dashed edges are contradictions.

Topics

projects
work
preferences
personal
health
relationships
general
Connection
Contradiction
Pinned ★

Click a node
to inspect it

Nodes0
Connections0
Contradictions0

THE COGNITIVE GAP

LLMs forget conversations when the context window resets. We fixed that.

One memory layer, every coding agent.

01Unified Memory

One store, every tool

Imprint sits beneath every coding agent and browser AI you use. Claude Code, Cursor, Codex, and Antigravity all draw from the same memory — your projects, preferences, and context travel with you, no matter which IDE you open.

One DynamoDB store · all tools share itSee how it connects →
02MCP · All IDEs

Every coding agent

Install the MCP server once. Register it with Claude Code, Cursor, Codex, Antigravity — or any IDE that speaks MCP. Set IMPRINT_PLATFORM and every agent silently recalls your full context at session start.

Claude Code · Cursor · Codex · Antigravity · CustomInstall MCP →

Get started

Works in every AI coding tool.

One MCP server. Persistent memory across Claude Code, Cursor, Codex, Antigravity — or any IDE you use.

1

Clone & install

git clone https://github.com/ayushraj-byte/Cognee-Imprint.git
cd Cognee-Imprint/mcp && npm install
2

Register with Claude Code

claude mcp add imprint --scope user \
  -- node /path/to/Cognee-Imprint/mcp/server.js
3

Or add manually to ~/.claude.json

{
  "mcpServers": {
    "imprint": {
      "command": "node",
      "args": [
        "/path/to/Cognee-Imprint/mcp/server.js"
      ],
      "env": {
        "IMPRINT_USER_ID": "your-user-id",
        "IMPRINT_API_BASE": "https://cognee-imprint.vercel.app",
        "IMPRINT_PLATFORM": "claude-code"
      }
    }
  }
}
4

Create ~/.claude/CLAUDE.md

Call get_memories at session start.
Call save_memory when you learn something.
Never announce you're doing this.
5

Verify connection

claude mcp list
# imprint  ✓ Connected
↓ Download ZIP~5 min setup

Your IDE not listed?

REAL-TIME EDGE SYNCHRONIZATION

As you work with your AI assistant, Imprint's background engine streams message vectors directly into an isolated AWS serverless database, creating a permanent recall layer.

Infrastructure x Intelligence.

Decoupled Architecture

By separating state from the raw model layer, Imprint makes your AI assistant natively stateful. We deploy specialized database nodes inside your private cloud, ensuring your operational context never enters public training loops.

Serverless Hydration

When a prompt is initialized, our Vercel Edge middleware performs a semantic vector lookup across AWS, injecting highly relevant historical schemas into the prompt context prior to model generation.

Why we built this

Your AI is brilliant.
It just forgets you exist.

The Problem

AI coding assistants forget everything the moment the context window resets. Every new session starts blind — no memory of who you are, what you're building, or how you think.

The Fix

Imprint captures lasting facts from every session, across every IDE, and feeds them through Cognee — a graph + vector memory engine that stores not just what you said, but how it all connects. Next session, your assistant already knows you.

The Edge

Because Cognee builds a knowledge graph, Imprint answers questions instead of dumping notes — and real-time contradiction detection flags when new facts conflict with your own memory. No other memory tool does this.

The Stack

Production-grade infrastructure, zero ops.

Cognee

Graph + vector memory engine

Next.js 16

App + API layer

Vercel Edge

Global deployment

AWS DynamoDB

Memory storage

Groq API

llama-3.3-70b · fast extraction

Jina Embeddings

1024-dim semantic retrieval

AES-256 BYOK

End-to-end key encryption

Core Pipeline

Knowledge Graph
The Memory Brain

Powered by Cognee

Every memory runs through Cognee’s add → cognify → search pipeline, building a knowledge graph alongside vector embeddings. Vectors surface what’s similar; the graph knows how your memories connect. With GRAPH_COMPLETION, Imprint doesn’t just hand back notes — it answers, grounded in everything it knows about you.

Memory Storage

AWS DynamoDB + Groq Extraction

Memories are extracted by Groq's llama-3.3-70b in real-time and stored in serverless DynamoDB — pinned facts, topic filters, and contradiction detection built in.

Session Management

Amazon DynamoDB Isolation

State storage utilizing low-latency serverless Amazon DynamoDB tables for instant execution. Complete user control over encryption keys and memory pruning.

Real-time Sync
Real-time Sync

Cross-IDE Memory Graph

Every memory saved — whether from Claude Code, Cursor, Codex, or Antigravity — lands in the same DynamoDB table. A live dashboard polls every 3 seconds, animating new memories as they arrive.

Semantic Retrieval
Semantic Retrieval

Jina Embeddings + Relevance Ranking

Jina AI embeds every memory at 1024 dimensions, so each session pulls the memories relevant to what you're asking — not just the most recent. Pinned facts are always injected, and duplicate saves are merged automatically.