alibaizhanov/mengram

View on GitHub ↗

Human-like memory for AI agents — semantic, episodic & procedural. Experience-driven procedures that learn from failures. Free API, Python & JS SDKs, LangChain, CrewAI & OpenClaw integrations.

189 ★29 forksPythonUpdated 21d ago

What you need to know

Memory API with SDKs that gives AI agents three types of learning memory - semantic facts, episodic events and procedures that evolve from failures

Install

pip install mengram-ai
mengram setup

Usage

  • m.add([...]) to store conversations, m.search() and m.ask() to retrieve with citations
  • Install Mengram for me by fetching the canonical install guide at https://mengram.io/agent-install.txt for full agent setup

Key features

  • Three memory types: semantic, episodic, and procedural with failure-driven evolution
  • Cognitive Profile generated from all memories for system-prompt personalization
  • Native multilingual retrieval across 23 languages
  • Ask & citations with synthesized grounded answers
  • Claude Code hooks for auto-save and auto-recall that survive /clear and compaction
  • Import Claude Code history, ChatGPT, and Obsidian vaults
  • Multi-user isolation and a 30-tool MCP server

Best for

Adding durable, evolving cross-session memory to AI coding agents and workflows

Caveats

  • Cloud edition needs an API key from mengram.io
  • Self-hosting with Ollama requires 8B+ parameter models with 8K+ context
  • Multilingual ask feature is premium (Pro/Growth/Business)
Clients: Claude Desktop · Claude Code · Cursor · Codex · Windsurf · Cline · OpenClaw

Reviewed 2026-08-11

Topics

agent-memoryai-agentsai-memoryclaude-desktopcognitive-architecturecoherecursor-aiepisodic-memoryknowledge-graphletta-alternativellm-memorymcp-servermem0-alternativemodel-context-protocolmultilingual-embeddingspgvectorprocedural-memorypythonragsemantic-search
Stars
189★
Forks
29
Language
Python
License
Apache-2.0
Created
2026-02-10
Last push
2026-08-14