engram

Store and recall persistent semantic memories for AI agents across sessions.

35|5|Updated Mar 28, 2026

Search Tags:#mcp #memory #semantic #qdrant #openclaw #three-tier #fastembed

Ready to run

npx skills add https://github.com/EngramMemory/engram-memory --skill engram-engrammemory

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: engram
Source: https://github.com/EngramMemory/engram-memory/tree/main
Command: npx skills add https://github.com/EngramMemory/engram-memory --skill engram-engrammemory

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires qdrant-client, requests, httpx, and includes scripts (resource) components.

What problem does it solve?

Engram provides persistent semantic memory for AI agents, enabling them to remember preferences, facts, and past decisions across sessions and to recall relevant context automatically.

Core Features & Use Cases

Auto-recall and auto-capture: inject memories into agent context and store important details after conversations.
Three-tier recall: hot cache, multi-head hash index, and vector search for fast, accurate retrieval.
Local-first with optional cloud sync: memories stay on-device with optional Engram Cloud for overflow and analytics.

Quick Start

Install Engram, deploy the all-in-one container, then store a memory and search memories.

Frequently Asked Questions about engram

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I add persistent semantic memory to an MCP-based AI agent?▼

Persistent semantic memory for MCP-based AI agents is added by deploying a local memory store that captures conversation context and injects relevant memories into future sessions automatically. Engram provides this using local Qdrant storage and FastEmbed embeddings.

Does semantic memory for AI agents work with local Qdrant storage?▼

Yes, semantic memory for AI agents works with local Qdrant storage to keep memories on-device. It uses FastEmbed embeddings and a three-tier recall pipeline consisting of a hot cache, multi-head hash index, and vector search for fast retrieval.

What's the best way to recall past context across different agent sessions?▼

The best way to recall past context across agent sessions is a three-tier recall pipeline combining a hot cache, multi-head hash index, and vector search. This approach retrieves relevant semantic memories fast and accurately without relying solely on linear history.

Can I use auto-recall and auto-capture to inject memories into AI conversations?▼

Yes, auto-recall and auto-capture can inject memories into AI conversations by automatically storing important details after interactions and fetching relevant context before the next response. This keeps agent context continuously updated across sessions.

How do I deploy a self-hosted memory store for AI agents?▼

To deploy a self-hosted memory store for AI agents, install the memory package and run an all-in-one container. This sets up local Qdrant storage with FastEmbed embeddings, allowing you to store and search memories immediately.

Do I need an internet connection to maintain local semantic memories for OpenClaw?▼

No, you do not need an internet connection to maintain local semantic memories for OpenClaw. Memories stay on-device using local Qdrant storage, with optional Engram Cloud integration available only for overflow and analytics.