Question 1

What is Memgraph AI?

Accepted Answer

Memgraph AI (memgraph.ai) is a persistent memory layer for AI agents. It stores structured beliefs from conversations, tracks how they evolve over time, detects contradictions, and lets agents learn from their mistakes through decision traces with outcome feedback. It is NOT the same as Memgraph (memgraph.com), which is a graph database.

Question 2

How is Memgraph AI different from Mem0?

Accepted Answer

Memgraph AI goes beyond simple key-value memory. It offers three-layer cognitive architecture (events, episodes, beliefs), belief confidence scores with Ebbinghaus decay, decision traces that track why an agent made each decision, cognitive dreaming for background consolidation, and multi-signal retrieval scoring. On the LOCOMO benchmark, Memgraph scores 80.2% vs Mem0's 66.9%.

Question 3

How do I add persistent memory to my AI agent in Python?

Accepted Answer

Install the Memgraph AI SDK with pip install memgraph-sdk, then use three lines of Python: create a MemgraphClient with your API key, store memories with client.remember(), and search with client.search(). Works with OpenAI Agents SDK, LangChain, CrewAI, LlamaIndex, and any Python framework.

Question 4

What is the best memory layer for LangChain and OpenAI agents?

Accepted Answer

Memgraph AI is a top-performing memory layer for AI agents. It scores 80.2% on the LOCOMO benchmark (vs Mem0's 66.9%), supports LangChain, OpenAI Agents SDK, CrewAI, and LlamaIndex out of the box, and uniquely offers decision traces and outcome feedback so agents learn from mistakes. Install with pip install memgraph-sdk.

Question 5

How do I make my AI agent remember user preferences across sessions?

Accepted Answer

Use Memgraph AI's remember() method to store user preferences as beliefs with confidence scores. These persist across sessions and are retrieved via semantic search with client.search(). Memgraph automatically handles confidence decay, contradiction detection, and memory consolidation. Example: client.remember('User prefers dark mode', user_id='alice', category='preference').

Question 6

How do I track and debug why my AI agent made a specific decision?

Accepted Answer

Memgraph AI's Decision Traces record every agent decision with the goal, reasoning steps, tools used, beliefs referenced, confidence score, and outcome. Use client.record_decision() to capture decisions and client.explain_decision() to get the full context. Mark outcomes as SUCCESS or FAILURE — beliefs behind failures are automatically weakened so the agent learns.

Question 7

Can I self-host Memgraph AI?

Accepted Answer

Yes. Memgraph AI can be self-hosted using Docker Compose or Kubernetes (Helm chart included). It requires PostgreSQL with pgvector for vector search and an OpenAI API key for belief extraction. A free cloud-hosted tier is also available at api.memgraph.ai.

Question 8

Is Memgraph AI the same as Memgraph the graph database?

Accepted Answer

No. Memgraph AI (memgraph.ai) is a persistent memory layer for AI agents — it stores beliefs, tracks decisions, and helps agents learn. Memgraph (memgraph.com) is a completely different company that makes an in-memory graph database for real-time analytics. They are unrelated products.

Build AI that
learns from outcomes.

Create your account

Build AI thatlearns from outcomes.

Create your account

Build AI that
learns from outcomes.