mie

Overview

MIE (Memory Intelligence Engine) creates a shared knowledge graph that AI agents like Claude, ChatGPT, and Cursor can access. It stores structured information as facts, decisions, entities, and events with relationships between them, rather than simple chat logs. The architecture uses an embedded CozoDB database with optional semantic search via Ollama embeddings. The MCP server exposes 12 tools for storing, querying, updating, and managing the knowledge graph, with zero server-side inference costs since it relies on the client agent to evaluate what's important.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• mie_analyze

  Surfaces related context before storing — the agent decides what's worth remembering

  Try:Show me related context before storing this informationTry:What should I remember about this decision?
• mie_store

  Writes facts, decisions, entities, events, and relationships to the graph

  Try:Store that we chose PostgreSQL over DynamoDB for ACID transactionsTry:Remember that Kraklabs is the independent software lab behind CIE
• mie_bulk_store

  Batch store up to 50 nodes with cross-references — ideal for importing knowledge from files or git history

  Try:Import all our decisions from these architecture documentsTry:Batch store these facts about our team and technologies
• mie_get

  Retrieve a single memory node by ID with full details

  Try:Get details about database decision node 123Try:Show me the full information about entity Kraklabs
• mie_query

  Semantic search, exact lookup, or graph traversal across all node types

  Try:Find all our security decisionsTry:Show me facts related to our payment architectureTry:What entities are connected to the PostgreSQL decision?
• mie_list

  List and filter nodes with pagination

  Try:List all events from last monthTry:Show me our active decisionsTry:List entities related to authentication
• mie_update

  Invalidate outdated facts, update statuses — with full history preserved

  Try:Update the database decision status to finalizedTry:Mark the old API fact as outdated
• mie_delete

  Remove nodes with cascade (embedding + edges) or remove individual relationships

  Try:Delete the outdated DynamoDB alternative from the PostgreSQL decisionTry:Remove the event about cancelled demo
• mie_conflicts

  Detect contradictions in stored knowledge

  Try:Check for conflicts in our security decisionsTry:Show me any contradictory information about database choices
• mie_export

  Export the full graph as JSON or Datalog

  Try:Export my entire memory graph as JSONTry:Create a backup of all knowledge
• mie_repair

  Rebuild HNSW indexes and clean orphaned embeddings

  Try:Rebuild the search indexesTry:Clean up orphaned embeddings in the graph
• mie_status

  Graph health, node counts, usage metrics

  Try:Show me graph statistics and healthTry:How many facts and decisions are stored?

Comparable tools

Installation

brew tap kraklabs/mie
brew install mie

mie init                    # Quick setup with defaults
mie init --interview        # Interactive setup

{
  "mcpServers": {
    "mie": {
      "command": "mie",
      "args": ["--mcp"]
    }
  }
}

FAQ

What makes MIE different from platform-specific memory features?

MIE works across all AI agents (Claude, ChatGPT, Cursor, etc.), stores structured data with relationships rather than flat text, and is portable/exportable. Platform memory is limited to a single provider and lacks rich relationships between knowledge items.

Can I use MIE without embeddings/Ollama?

Yes, MIE works without embeddings. You'll still get graph traversal and exact-match search capabilities. Embeddings only add semantic search capabilities.

On Hacker News

Recent discussion from the developer community.

• Show HN: MIE – Shared memory for all your AI agents (Claude, Cursor, ChatGPT)▲ 1
  Story by kraklabs · 2026-02-06

Compare mie with