Zengram FAQs

Question 1

What is Zengram?

Accepted Answer

Zengram is a shared memory system specifically designed for multi-agent AI applications. It enables seamless knowledge retention and collaboration across diverse AI agents, systems, and machines, preventing agents from forgetting crucial information.

Question 2

What is Zengram's 'Typed Memory System'?

Accepted Answer

The Typed Memory System allows you to manage different types of information – Events (immutable history), Facts (upsert by key), Statuses (current state), and Decisions (choices and reasoning). Each type has distinct lifecycles, decay rules, and mutation semantics.

Question 3

Is Zengram open-source and self-hostable?

Accepted Answer

Yes, Zengram is fully open-source with an MIT License. It is designed for self-hosting via Docker and provides official SDKs for Python and integrations for tools like Claude Code (MCP Server), n8n, and OpenClaw.

Question 4

What are the key benefits for multi-agent AI systems?

Accepted Answer

Zengram offers cross-agent briefings, corroboration (two agents storing the same fact strengthens its validity), agent-scoped API keys, and automatic credential scrubbing. It also uses LLM Consolidation to merge duplicates, flag contradictions, and discover new connections.

Question 5

How does Zengram store and retrieve memories?

Accepted Answer

Zengram uses a Dual Storage Architecture (Qdrant for semantic vector search and SQL databases for structured queries) and a Multi-Path Retrieval system. It combines vector, keyword (BM25), and entity graph searches using Reciprocal Rank Fusion for highly accurate retrieval.

Zengram

Zengram

主要功能

使用案例

主要功能

使用案例