# Mengram

> Open source AI memory API with human-like architecture featuring semantic, episodic, and procedural memory types for personalized AI experiences.

Mengram provides an AI memory API designed with human-like cognitive architecture, offering three distinct memory types: semantic (facts and preferences), episodic (events and decisions), and procedural (learned workflows). It enables developers to replace complex RAG pipelines with a single API call that returns a Cognitive Profile—a ready-to-use system prompt for instant AI personalization.

- **Three Memory Types** — Stores semantic memories for facts and preferences, episodic memories for events and discussions, and procedural memories for workflows and habits, mimicking how the human brain organizes information.

- **Cognitive Profile** — Generates a unified system prompt from all memory types with one API call, enabling instant personalization for any LLM without complex retrieval setups.

- **Memory Agents** — Includes autonomous agents: Curator cleans contradictions, Connector finds hidden patterns, and Digest provides weekly briefs automatically.

- **Unified Search** — Searches across all three memory types simultaneously using vector search, BM25, graph expansion, and LLM re-ranking in a single call.

- **Procedure Feedback** — Tracks success and failure counts per workflow, allowing AI to learn which procedures work best and surface proven patterns first.

- **Experience-Driven Procedures** — Self-improving workflows that auto-evolve when failures are reported and auto-create new procedures from repeated successes.

- **Smart Triggers** — Proactive memory features including reminder detection from conversations, contradiction alerts, and workflow pattern recognition.

- **Knowledge Graph** — Stores entities and relations as structured data rather than plain text, enabling precise relationship tracking.

- **Team Memory** — Enables shared knowledge across team members with simple invite codes for quick onboarding.

- **Import Capabilities** — Supports importing existing data from ChatGPT exports, Obsidian vaults, and text files to eliminate cold-start problems.

- **Framework Integrations** — Drop-in support for LangChain, CrewAI, and OpenClaw with dedicated memory classes and tools.

- **MCP Server** — Native integration with Claude Desktop, Cursor, Windsurf, and other MCP-compatible AI tools.

To get started, install via pip (`pip install mengram-ai`) or npm (`npm install mengram-ai`), obtain a free API key from the dashboard, and configure your preferred AI tool or framework. The SDK supports both Python and JavaScript with full TypeScript definitions.

## Features
- Three memory types: semantic, episodic, and procedural
- Cognitive Profile generation
- Autonomous memory agents (Curator, Connector, Digest)
- Unified search across all memory types
- Procedure feedback and success tracking
- Experience-driven procedure evolution
- Smart triggers and proactive alerts
- Knowledge graph with entities and relations
- AI reflections and insights
- Team shared memory
- Webhooks for memory change notifications
- Import from ChatGPT, Obsidian, and text files
- MCP server support
- Python and JavaScript SDKs
- Self-hostable option

## Integrations
Claude Desktop, Cursor, Windsurf, ChatGPT, LangChain, CrewAI, OpenClaw, Perplexity, Slack, Zapier, Notion

## Platforms
WEB, API, DEVELOPER_SDK

## Pricing
Open Source

## Version
2.9.0

## Links
- Website: https://mengram.io
- Documentation: https://mengram.io/docs
- Repository: https://github.com/alibaizhanov/mengram
- EveryDev.ai: https://www.everydev.ai/tools/mengram