Agent Memory vs. memory middleware

Q: Can I use LangChain or CrewAI with SurrealDB instead of their built-in memory?

Yes. SurrealDB integrates with LangChain, LlamaIndex and any MCP-compatible client, so a framework's memory module can be swapped for Agent Memory without changing the orchestration logic.

Q: Does removing middleware reduce latency?

Yes. Memory middleware adds a network hop to every memory operation. With Agent Memory, reads and writes happen inside the database: one round trip instead of two, which compounds under load and across multi-agent workflows.

Explore Agent Memory Join the Agent Memory waitlist

Middleware. Mem0, Zep and Letta sit between the agent and its storage.

The cost. Each layer adds latency, cost and a consistency gap.

Agent Memory. Memory in the database itself, one transaction from the data.

THE LANDSCAPE

A category with one shared limit

Mem0

LLM-driven extraction and consolidation, with optional graph storage. Strong developer experience over a set of separate stores.

Zep

Long-term memory with temporal knowledge graphs and entity extraction. The structure arrives as a separate service with its own consistency model.

Letta (MemGPT)

A self-editing memory framework where the agent manages its own context window. Memory sits outside the transactional boundary.

THE COMPARISON

A layer between the agent and its data. What the middleware adds, and what the database already holds.

Consistency

Agent Memory

Memory middleware

Atomicity

One transactionMemory and state commit together in the database.

Two systemsMemory commits in the middleware, state in the database, separately.

Round trips

One hopThe agent reads and writes memory straight in the database.

Two hopsEvery memory operation goes through the service, then the store.

Operations

Services to run

The databaseMemory is a set of tables and indexes in the database you already run.

One more serviceA memory service to deploy, scale and secure beside the database.

Permissions

One modelRBAC and record-level permissions cover memory and data alike.

Two modelsOne permission model for memory, another for the data it refers to.

THE ALTERNATIVE

Memory inside the database

Transactions

ACID memory

Memory and application state commit together, so they cannot diverge.

How transactions work

Permissions

One permission model

RBAC and record-level permissions apply to memory and data alike.

Explore permissions

Round trips

One round trip

Agents write straight to the database, in the same round trip as the read.

Explore Agent Memory

HOW IT WORKS

Recall and remember in one round trip. The agent reads its memory and writes the outcome in the same transaction.

1BEGIN TRANSACTION;
2
3-- Recall what the agent knows about this user
4LET $context = SELECT fact, valid_from FROM memory
5    WHERE agent = agent:support AND user = user:jaime
6    ORDER BY valid_from DESC LIMIT 5;
7
8-- Remember the outcome, beside the data it concerns
9CREATE memory SET
10    agent = agent:support,
11    user = user:jaime,
12    fact = "Moved to the Memory+ plan",
13    source = conversation:4471,
14    valid_from = time::now();
15
16COMMIT TRANSACTION;

TRUSTED BY

Enterprise teams building on SurrealDB

See all case studies

Three data stores collapsed into one

An AI assistant for 10,000 technicians

Nine backend tools consolidated into one

Low-latency RAG, under the customer's control

45 million recommendation queries a month

Memory middleware,answered

How does Agent Memory differ from Mem0 or Zep?

Mem0 and Zep are middleware layers between the agent and its storage, abstracting over a vector store, a document store and a graph engine that stay separate. Agent Memory is built into SurrealDB, so memory and application data share one transactional store with no consistency gap between them.

Can I use LangChain or CrewAI with SurrealDB instead of their built-in memory?

Does removing middleware reduce latency?

THE PLATFORM

Everything an application and its agents know. Five surfaces, one engine.

Database

Document, graph, vector, time-series and relational in one engine.

Five data models as dotted tiles: documents, graph, vector, time-series and relational

Agent Memory

What an agent learns, with its source and its time, in the same engine.

A timeline of remembered facts, each with its source

Cloud

Managed clusters in the regions you choose, scaled on demand.

Clusters in three regions on a world map, each running or scaling

Studio

Query, explore and design the schema from the browser.

MCP

Every model that speaks MCP reaches the database and the memory directly.

Three models connected through MCP to the database and Agent Memory

IN PRODUCTION

Trusted at scale. Samsung, Nvidia, Verizon, Tencent, and Walmart run on SurrealDB.

14,000+

Developers building on SurrealDB Cloud

4M+

Developers building on SurrealDB worldwide

FROM THE TEAMS

SurrealDB gives us a foundation where we can unify semantic search, knowledge graphs, and AI-driven decision making without stitching together multiple systems. Collapsing responsibility into SurrealDB has become our default engineering posture.

Justin Foley

VP of Engineering, Later

GET STARTED

Memory without the middleware. Agent Memory keeps memory in the database: one transaction, one permission model, one service to run.

Explore Agent Memory Join the Agent Memory waitlist