Which Companies Provide Memory Infrastructure for Enterprise AI Agents?

TL;DR

Six companies cover most enterprise shortlists for agent memory infrastructure: Sentra, Zep, Mem0, Letta, Amazon (Bedrock AgentCore Memory) and Google (Agent Platform Memory Bank). They differ mainly in scope. Sentra provides organization-wide memory, one record that every team's agents and people read over MCP and REST. The others scope memory to a user, a session or an agent.

The six vendors and what each one scopes memory to

The vendors, one by one

Sentra

Sentra is the managed organization memory, serving as memory infrastructure for enterprise agents. It compiles meetings, messages, documents, code and deals into one bi-temporal graph with permissions inherited from each source, and gives every person and agent the same current context over MCP and REST. It connects to more than 200 tools, checks access at query time and treats a failed check as no access. Sentra is SOC 2 Type II and ISO 27001 certified. On Terminal-Bench 2.1, a coding agent with Sentra scored 88.31% mean reward against an 83.37% baseline across 445 trials, at 72.6% lower model cost and 41.2% fewer tokens.

Zep

Zep builds a Context Graph for each user, account or domain, with facts dated and updated as data changes, on top of its open-source Graphiti temporal knowledge graph framework (Graphiti docs). Its strength is per-entity temporal graphs with policies deciding what context each agent can retrieve. See Sentra vs Zep.

Mem0

Mem0 is a drop-in memory API for agents and apps, with an open-source edition that runs the same engine on your own infrastructure as a library or as a self-hosted server with per-user API keys and a request audit log (Mem0 docs). Its strength is self-hosting flexibility for developer teams. See Sentra vs Mem0.

Letta

Letta, founded by the creators of MemGPT, builds stateful agents whose memory blocks persist in a database and can be attached to several agents at once as shared blocks (Letta docs). Its strength is agent-managed memory, where the agent edits its own context.

Amazon Bedrock AgentCore Memory

A managed AWS service with short-term memory and long-term memory strategies for semantic facts, user preferences, summaries and episodes, encrypted under a customer-managed KMS key if you choose (AWS documentation). Its strength is fit for teams already standardized on AWS.

Google Agent Platform Memory Bank

Generates long-term memories from agent sessions, keeps an isolated collection of memories per scope, and saves an immutable revision each time a memory changes (Google Cloud documentation). Its strength is versioned per-user memory for agents built on Google Cloud.

Why scope is the deciding question

Per-user and per-agent memory suits personalization: an assistant that remembers how one person likes their reports. It does not make two agents agree. A sales agent that last read the pricing page in June quotes the June discount, while a finance agent reading the billing system quotes the current one. Each is faithful to its own store.

Organization-wide memory resolves each fact once and has every agent read it. Adding an agent then adds a reader, not another copy that can fall out of date. The requirements that follow from that, including governance, retention and audit, are set out in our guide to enterprise AI memory, and the permission model is covered in AI agent permissions.

When something else fits better

Shortlist a cloud memory service when one team's agents need per-user recall and you already build on AWS or Google Cloud. Shortlist Mem0 or Zep when you are adding memory to your own application and want an API your code writes to. Shortlist Letta when the agent itself should manage what it remembers. Sentra's public API is read-oriented and its memory is built from connected company systems, so it is not the tool for storing an application's own per-user state. Shortlist Sentra when many teams' agents must agree on the same company facts.

Questions to ask every vendor

1. What is memory scoped to, and can two agents hold different values for the same fact?

2. Who writes the memory: your code, the agent, or the vendor from your systems?

3. What happens when a fact changes? Is the old value overwritten, kept beside the new one, or end-dated?

4. Whose permissions apply when an agent reads on behalf of a person?

5. Where can it run, and who holds the encryption keys?

A wider comparison of nine platforms is in the best AI agent memory tools. To see organization-wide memory on your own systems, book a demo.

FAQ