Agent Memory FAQs

Question 1

How does Agent Memory ensure accurate and relevant search results?

Accepted Answer

It employs a hybrid search mechanism, combining vector similarity (semantic search) with BM25 keyword search. This powerful combination ensures optimal recall, effectively finding both conceptually similar information and exact term matches within an agent's memory.

Question 2

What is Agent Memory and how does it benefit AI agents?

Accepted Answer

Agent Memory is a Model Context Protocol (MCP) server that provides AI agents with persistent, searchable long-term memory using PostgreSQL and pgvector. It allows agents to recall facts, prompts, notes, and documents across sessions, overcoming their default statelessness and enabling more intelligent, context-aware interactions.

Question 3

Can Agent Memory be used for multiple AI agents or users?

Accepted Answer

Yes, Agent Memory is designed with multi-tenancy support, allowing you to isolate memories per agent, user, or deployment using `tenant_id`. Additionally, `namespaces` enable further subdivision of memories within a tenant, ensuring data privacy and organization.

Question 4

What types of information can Agent Memory store?

Accepted Answer

Agent Memory supports four distinct types: `FACT` for discrete ground-truth knowledge, `PROMPT` for reusable system instructions, `NOTE` for general episodic memories, and `DOCUMENT` for large text documents which are automatically chunked for efficient storage and retrieval.

Question 5

What are the technical requirements to use Agent Memory?

Accepted Answer

You'll need Java 17+, Maven 3.8+, and PostgreSQL 14+ with the pgvector extension. For embeddings, you can choose between an OpenAI API key or a local Ollama instance running the `nomic-embed-text` model.

Agent Memory

Agent Memory

主要功能

使用案例

主要功能

使用案例