
What is a company brain for AI (and what it really ingests today)
A company brain is the memory your AI agents query, with citations you can verify. What it is, which sources it ingests today, who it suits and what it is not.

Agent memory layers, assistant memory, enterprise search and DIY frameworks: public pricing checked on 2026-11-14 and which option fits which kind of team.
If you searched for "AI memory" this year, you will have seen products that look nothing alike: an API that lets your agent remember its users, a toggle in your assistant that saves your preferences, a search engine that indexes the whole company, and a pile of libraries to build it yourself. They all use the word memory. They solve different problems and are paid for in very different ways.
This map sorts the market into categories, explains how each one charges using public information from their websites checked on 2026-11-14, and says who each option suits. We are an interested party: we build MemoryFirst. So you will see what we verified and what we did not.
| Category | Examples | What it remembers | Who uses it |
|---|---|---|---|
| Memory layer for agents | Mem0, Zep, Letta, Supermemory | What an agent you build needs to know about its users or tasks | Teams developing products with agents |
| Memory built into assistants | ChatGPT, Claude | Each person's preferences and context inside that assistant | Anyone using that assistant |
| Enterprise search | Glean | The company's documents and apps, with permissions | Large companies with many tools |
| Frameworks and in-house parts | Open-source libraries, Postgres with vectors | Whatever you program | Teams with engineering to spare |
MemoryFirst sits between the first and the third: it is a memory layer your agents consume over MCP, but aimed at a team's or company's knowledge (decisions, customers, deadlines), with citations. More on that below.

Plans and prices change fast. Everything below comes from the public pages cited under "Sources", checked on 2026-11-14. Where a detail did not appear, we say so.
| Product | What its public site shows | What is not visible |
|---|---|---|
| Mem0 | A Free plan (10,000 adds and 1,000 retrievals a month for the first 30 days; 50,000 and 10,000 with your own LLM key) and a Custom plan with price on request, adding graph memory, SSO, audit logs, SLA and on-prem deployment. An open-source library under Apache 2.0 | Intermediate prices. Our July comparison listed $19 and $249 tiers that do not appear on the page today: verify the current ones |
| Zep | Free with 10,000 credits a month; Flex at $125 a month (50,000 credits); Flex Plus at $375 a month (200,000); custom Enterprise with SOC 2 Type II, a HIPAA BAA and options to deploy with your own keys or in your cloud | Annual price (the page mentions 17% savings with no figures). Graphiti, its graph engine, is open source under Apache-2.0 |
| Letta | Free at $0 (up to 3 stateful agents), Pro at $20 a month with $20 in credits, Teams Pro at $20 per seat and custom Enterprise | Letta's pricing pages change often and gave us different versions: we could not verify other plans that appear on third-party sites |
| Supermemory | Free ($5 in monthly credits), Pro at $19 a month (3 seats), Max at $100, Scale at $399 (SOC 2, HIPAA BAA, self-hosted option) and custom Enterprise. The engine is not open source; the main repository, the MCP server and the plugins are (MIT) | Explicit usage limits per plan |
| Glean | Publishes no prices: the button is "Get a demo" | Any figure. The per-user estimate that circulates is one we could not check |
| ChatGPT and Claude | Memory inside each assistant, managed by the person. In Claude, each project has its own memory and, on Team and Enterprise, the owner switches the feature on or off without being able to see members' memories | What each ChatGPT plan includes: OpenAI's help page could not be opened, and we claim no more than what we saw in search results |
| Frameworks | Libraries and databases under open licences. The cost is yours: engineering and operations | We did not verify prices or licences of specific frameworks beyond Mem0 and Graphiti |
Two useful readings. First: the developer layers charge by usage (credits, adds, retrievals), and the cost grows with your product. Second: what a plan really costs often sits in the security fine print (SOC 2, HIPAA, deployment in your cloud), reserved almost always for the custom or most expensive plan.
Agent layers (Mem0, Zep, Letta, Supermemory). If you are building a product with agents and each user needs memory, this is your category. Zep stands out for temporal graphs with an SDK; Mem0 publishes its library under an open licence alongside the managed service; Letta offers stateful agents; Supermemory offers plugins for Claude Code and Cursor and a $19 entry plan. It is the best fit when you want to control how it integrates into your code.
Assistant memory. If you work alone, with one assistant, and it is enough that it remembers your preferences, you already have it. The limits: it lives inside that assistant (what Claude knows, your agent in Cursor does not) and it belongs to one person, not a team.
Enterprise search (Glean). If your problem is that hundreds of people cannot find documents scattered across dozens of applications, with per-user permissions, this is the right category. Glean says it has more than 275 connectors and permission-enforced search. To see a price you have to talk to sales.
Frameworks. If your team has engineering to spare and memory is part of your competitive edge, building it makes sense. If not, you are paying in hours what others charge as a subscription, and maintaining a memory system (quotas, deletion, audit, isolation) costs more than it looks on the first afternoon.

MemoryFirst is a managed memory for a team's knowledge, consumed by your agents and assistants over MCP (Claude, ChatGPT, Cursor and VS Code), with hooks for Claude Code. Every answer cites its source, and contradicting facts resolve toward the most recent one.
Where we differ: we think about teams that use ready-made agents, not people building an agent, and our answer includes the citation. Where we lose: if you need a deep SDK, the most flexible temporal graphs, deployment in your cloud or published certifications, the other options are more mature on those pieces.

We wrote each comparison separately, with pros and cons of both sides and a "when to choose the other" section:
For the reasoning behind all this, read why AI agents need a memory layer. And to see it applied, look at the software team use case and the one for an agency with many clients, both simulated scenarios.
All accessed 2026-11-14.
Start free with one project and compare it against whatever you use today, with your own data.
Start free
A company brain is the memory your AI agents query, with citations you can verify. What it is, which sources it ingests today, who it suits and what it is not.

What it costs when an AI agent starts each session with no past: minutes, tokens and reopened decisions, with an illustrative sum and the market options.

The short story of MemoryFirst: the problem we saw, why we built a memory layer with verifiable citations, and why we decided to build it ourselves, in Spain.