OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Agent Memory & RAG

Engram Memory

Engram Memory is a self-hosted, three-tier memory system for AI agents. It packages Qdrant, FastEmbed and an MCP server in a Docker container, letting an agent store, search, recall and forget memories across sessions using semantic embeddings, hash-based candidate retrieval and hybrid vector search.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Engram Memory — OpenViST project profile in Agent Memory & RAG.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Engram when an AI agent needs to retain information between sessions and the workflow can connect to its MCP server or another documented transport. The local Community setup is suited to operators who want the memory engine and vectors on their own hardware; cloud extensions require an API key and send text to Engram's API for specified processing.

Who it is for

  • Developers adding cross-session memory to an AI agent through MCP or another documented transport.
  • Operators who want a self-hosted memory service with embeddings and vectors kept on their own machine by default.
  • Teams that need tools for searching, consolidating, connecting and sharing agent memories.

Documented uses

  • Persist agent memories across sessions.
  • Search for relevant stored information and inject recalled memories into agent context.
  • Remove, retrieve, browse, consolidate or connect memory records.
  • Manage shared memory pools and access grants across devices.
  • Generate a visual graph of memories through the documented Claude Code command.
Before you choose

Requirements & limits

The repository README identifies a BSL 1.1 license. It describes the Community edition as free for internal use, research and self-hosted deployments; commercial or SaaS use requires a license, and each version converts to Apache 2.0 after four years. The self-hosted setup requires Docker, 4 GB or more of RAM and 10 GB or more of storage. Python 3.10 or later is needed only to run the stdio MCP server or CLI tools di

  • The self-hosted container bundles Qdrant, FastEmbed and the MCP server; embeddings and vectors run locally by default.
  • Community retrieval combines a hot cache, locality-sensitive hashing and hybrid dense-plus-sparse search rather than relying on LLM reranking.
  • Community limits include 1,000 hot-cache entries, 500 graph entities with one-hop traversal, and manual deduplication at a fixed 0.95 threshold.
  • With an Engram Cloud API key configured, text is sent to the cloud API for compression, deduplication and category detection; vectors remain in the local Qdrant store, according to the README.
  • The README lists MCP, SSE, stdio and REST transports, so connection method depends on the agent client or integration.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up Engram Memory and connect an MCP client

Start the self-hosted container and register it with an MCP client so the client can reach Engram's memory tools.

Before you start

  • Docker
  • At least 4 GB RAM and 10 GB storage
  • An MCP-compatible client; the documented example uses Claude Code

Steps

  1. Start the container with the documented Docker command, exposing ports 6333, 11435 and 8585 and mounting the engram_data volume at /data.

  2. For Claude Code, register the server with claude mcp add engrammemory -s user --transport http http://localhost:8585/mcp.

  3. Use the client's MCP connection to access Engram's memory tools.

Expected outcome

A running local Engram container with persistent storage and an MCP endpoint registered for the client.

Guide limitations

  • This covers setup and connection, not a verified execution test.
  • The Docker command and Claude Code registration are the documented path; other clients have their own connection steps.
  • The Community edition has documented capacity and feature limits.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Vector search

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
Engram MemoryPersist agent memories across sessions.Linux, macOS, Self-hosted, WindowsThe repository README identifies a BSL 1.1 license. It describes the Community edition as free for internal use, research and self-hosted deployments; commercial or SaaS use requires a license, and each version converts to Apache 2.0 after four years. The self-hosted setup requires Docker, 4 GB or more of RAM and 10 GB or more of storage. Python 3.10 or later is needed only to run the stdio MCP server or CLI tools di
YourmemoryRetain durable facts and preferences across agent sessions.CLI, Self-hosted, ServerThe Python package setup requires obtaining a token through the project website by entering an email and verifying a six-digit code. The README also describes standalone binaries for macOS, Linux and Windows.
ContextatlasIndex a repository and find relevant implementation details using a natural-language question.CLI, ServerThe project is open source under the MIT license. Its documented requirements include Node.js 20 or later and embedding and reranking API credentials and settings.
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
EngramMemory/engram-memory · Official homepage title · retained in its source language
How we review sourcesSuggest a correction