
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.

An honest Zep and Graphiti vs MemoryFirst comparison: verified pricing and credits, MCP, temporal graph, compliance and when each fits. Data of 2026-11-14.
Anyone who has looked at the market for AI agent memory has run into two names that sound alike: Graphiti and Zep. Graphiti is the open-source graph engine; Zep is the company and the managed service that uses it inside. We are MemoryFirst and we compete with Zep in some cases, so we are an interested party. What follows tries to be fair: every Zep fact has its source and date at the end, and what we could not check, we say so.
The dividing line: Zep and Graphiti are better if you build agents and need to control the graph; MemoryFirst is better if you do not want to build anything. And on compliance and enterprise deployment, Zep currently publishes more options than we do.
| Aspect | Zep (and Graphiti) | MemoryFirst |
|---|---|---|
| What it is | Context layer with temporal graphs, SDKs and MCP; Graphiti is its open engine | Memory service to plug into agents over MCP and REST |
| Open source | Graphiti: Apache 2.0. Zep Cloud: managed service. Zep Community Edition is deprecated and unsupported | No |
| Self-hosting | Graphiti, yes (you provide the graph database). Zep in your cloud (BYOC): Enterprise plan only, via sales | No self-hosted edition |
| MCP | Their pages mention MCP access and "seats" for the context MCP server in each plan; we did not verify its tools. Graphiti's reference server warns it has no authentication | Remote at mcp.memoryfirst.ai with 15 tools and OAuth 2.1 or API key, plus a stdio server |
| Temporal graph | Their specialty (temporal context graphs) | Temporal graph with citations and resolved contradictions on Pro |
| Citations to the origin | They say you can "trace graph data to its sources"; we did not verify the citation format | Every answer cites its source (page, minute, message, URL) |
| Connectors | Google Drive, Slack, Notion, Gmail and Granola from Flex; more (Teams, SharePoint, Snowflake, Databricks, BigQuery) on Enterprise | Gmail and IMAP, Calendar, Drive, Dropbox, OneDrive, Slack, Telegram, files and events through the API |
| Where data is hosted | Their pages give no regions. They say they sign data processing agreements with EU customers | Database, embeddings and files in the EU (Germany) |
| Certifications | SOC 2 Type II and HIPAA BAA: advertised, although their plan table marks them as Enterprise only | GDPR by design; we do not claim certifications we do not hold |
Zep, as of 2026-11-14 (prices in US dollars, billed monthly):
A credit corresponds to an "episode", any object you send (a message, a JSON payload, a block of text): up to 350 bytes costs 1 credit and each extra 350-byte chunk costs one more. According to their page, retrieval, storage and users do not consume credits.

Simulated scenario: a fictional startup sends 2,000 episodes a day of about 700 bytes each. Each episode costs 2 credits, so that is 4,000 credits a day, around 120,000 a month. That is well above Flex's 50,000 and fits inside Flex Plus's 200,000. This is an illustrative calculation with invented numbers, not a real case.
MemoryFirst (from our pricing page, on the date of this article): Free (0 euros, 1 project, 1 seat, 30-day retention); Pro, 12 euros a month billed annually (144 euros a year) or 14.40 paid monthly, with unlimited projects and retention, all connectors and a temporal graph with citations; Team, 39 euros a month billed annually (468 euros a year) or 46.80 monthly, with 5 seats and shared memory. Pioneer offer: Pro for 9 euros a month for life, for the first 50 seats and until 31 December 2026.
The difference in model matters more than the number. With Zep, cost grows with the volume of episodes you send; with MemoryFirst you pay for a plan with seats and connectors, and quotas apply per plan. We are not comparing $125 with 12 euros as if they were the same thing.
Pros
Cons
Pros
Cons

It wins on technical depth of the graph, on enterprise options (own keys, customer cloud, certifications and a BAA, all as they publish them) and on developer control: SDKs in three languages and an open framework. If your project is an agent product and the graph is the heart of it, it is a serious choice. And if your organization requires deployment in your own cloud, we cannot offer that today.

Keep reading. The full context is in the AI memory market in 2026. For a joint view of Mem0, Supermemory and Zep, read how to choose a memory layer; for the other comparison in this series, MemoryFirst vs Mem0. You can also read how we retrieve information, citations and GDPR, the pricing page and the short comparison.
All accessed 2026-11-14. Anything not listed here, or marked "we could not verify", is not claimed.
The Free plan includes one project, MCP memory and the Claude Code hooks, with data stored in the EU.
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.

Mem0, Supermemory, Zep, MemoryFirst. They all promise 'memory for AI'. The differences that actually matter: where your data lives, whether you can audit answers, and which model reads your text.