Engram

What is Engram?

Engram is a universal AI memory backend that gives any AI model a persistent, growing brain. It stores memories in a local SQLite database with full-text search, vector semantic search, and a knowledge graph that connects related concepts.

How to use Engram?

Install the server with npx -y @engram-ai-memory/mcp@latest. Configure it as an MCP server in any compatible client such as Claude Desktop by adding the command and arguments to the mcpServers section of your client’s configuration.

Key features of Engram

• Semantic search finds memories by meaning, not just keywords
• 7-step knowledge graph traversal connects related memories
• Contradiction detection automatically flags conflicting beliefs
• Local-first – all embeddings run on-device, no API keys needed
• Universal – works with Claude, OpenAI, Ollama, and any MCP client
• 18 built-in MCP tools for memory management

Use cases of Engram

• Storing and retrieving past conversations, decisions, and events (episodic memory)
• Maintaining a persistent knowledge base of facts and beliefs (semantic memory)
• Recording how-to instructions, workflows, and code patterns (procedural memory)
• Detecting and resolving contradictory information in an AI’s knowledge base
• Providing long-term context to AI models across sessions and clients

FAQ from Engram

What types of memory does Engram support?

Engram supports three memory types: episodic (past events and conversations), semantic (facts and beliefs), and procedural (how-to instructions and workflows).

Does Engram require API keys or cloud services?

No. Engram is local-first – all embedding and search run on-device, and no data leaves your machine. No API keys are needed.

Which AI clients can use Engram?

Engram works with Claude, OpenAI, Ollama, and any MCP-compatible client.

Where are memories stored?

Memories are stored in a local SQLite database on your machine.

What MCP tools does Engram provide?

Engram offers 18 tools, including store_memory, search_memory, recall_context, add_knowledge, check_contradictions, resolve_contradiction, forget, tag_memory, list_tags, decay_sweep, decay_policy, re_embed, embedding_status, index_status, memory_stats, plugin_list, webhook_subscribe, and webhook_list.