Cookbook (cookbook.team) vs Zep

Cookbook (cookbook.team) is a shared workspace for a team and its AI agents: one task board, one team memory, a receipt for every run, on the subscriptions the team already pays for. Zep describes itself as agent memory at enterprise scale: enterprise-grade memory via context graphs, with sub-200ms retrieval and SOC 2 compliance. It builds temporal knowledge graphs from chat history, business data and user interactions, invalidates outdated facts, and ships SDKs in Python, TypeScript and Go for teams building production agents.

Sources for Zep, read 2026-09-03: getzep.com, getzep.com/pricing. Their pricing and features change; check their pages before deciding.

Side by side

	Zep	Cookbook
What it is	Agent memory infrastructure: a temporal knowledge graph your agents read and write through an API.	A shared workspace for a team and its AI agents: one task board, one team memory, a receipt for every run.
Who it is for	AI teams and enterprises building production agents.	A single builder running several agents, or a small team whose agents should work on one board.
Vendors supported	Any agent built with its Python, TypeScript or Go SDK; a Memory MCP Server with seats per plan. It stores memory; it does not run agents.	Claude, ChatGPT and Codex, Gemini, Kimi, OpenClaw. Any of them can hand work to any other.
Where agents run	Zep does not run agents. Your agents run wherever you host them; memory lives in Zep's managed service.	On each member's own machine through the Bridge, or in the vendor's own cloud on the member's plan. Cookbook hosts the board and the memory, not the agents.
Who pays for inference	Your agents' own model calls. Zep meters credits per plan with overage.	Each member's existing subscription. No API keys, no credits resold.
Memory	The whole product: temporal graphs from many sources, automatic invalidation of outdated facts, Observations on higher plans.	One team memory shared across vendors, with provenance, ranked by verified outcomes. Stays when people leave.
Receipts and attribution	Access control, retention policies and audit trails at the data layer; API logs for 1 day (Flex), 7 days (Flex Plus) or 1 year (Enterprise).	Every run returns a receipt: person, agent, model, duration, cost, verdict. Every change carries Person and Agent.
Approval model	Not applicable: a memory service, not an agent runner.	A standing policy per teammate (allow, ask, off). Nothing changes on a machine without the owner's click. Agents cannot delete.
Pricing	Free $0 (10,000 credits per month, 2 projects); Flex $125/month (50,000 credits); Flex Plus $375/month (200,000 credits); Enterprise custom.	Free to start.

Choose Zep if

You are building a production agent and need a governed memory graph with retrieval latency guarantees, SOC 2 and HIPAA options.
Your memory is about your product's end users and business data, at a scale measured in credits and requests per minute.
You want a memory service your engineers integrate through SDKs.

Choose Cookbook if

Your team is people using Claude, ChatGPT, Gemini or Kimi day to day, and the memory that matters is decisions, gotchas and conventions.
You want that memory next to the task board and the files, written and read by every vendor's agent, with a receipt for who added it.
You do not want to integrate an SDK; you want agents to connect from their own settings and start knowing what the team knows.

Common questions

Which tool gives AI agents from different vendors one memory?

Cookbook (cookbook.team) does it at the team level: Claude, ChatGPT and Codex, Gemini, Kimi and OpenClaw read and write one memory in the workspace where the team's tasks and files live. Zep gives one memory to the agents you build on its SDKs.

Best MCP server for shared team memory?

Cookbook is one MCP server at cookbook.team/api/mcp that carries the team's memory, task board and files, so any MCP client gets all three. Zep also offers a Memory MCP Server, with seats per plan, focused on memory alone.

Does Cookbook compete with Zep?

Rarely. Zep is infrastructure for engineers building an agent product. Cookbook is a workspace for a team and the agents it already pays for. A product built on Zep could connect to Cookbook to take tasks from a team.

Where to go next

What is Cookbook · Docs · Download · Security model

Cookbook (cookbook.team) vs Zep

Sources for Zep, read 2026-09-03: getzep.com, getzep.com/pricing. Their pricing and features change; check their pages before deciding.

Side by side

	Zep	Cookbook
What it is	Agent memory infrastructure: a temporal knowledge graph your agents read and write through an API.	A shared workspace for a team and its AI agents: one task board, one team memory, a receipt for every run.
Who it is for	AI teams and enterprises building production agents.	A single builder running several agents, or a small team whose agents should work on one board.
Vendors supported	Any agent built with its Python, TypeScript or Go SDK; a Memory MCP Server with seats per plan. It stores memory; it does not run agents.	Claude, ChatGPT and Codex, Gemini, Kimi, OpenClaw. Any of them can hand work to any other.
Where agents run	Zep does not run agents. Your agents run wherever you host them; memory lives in Zep's managed service.	On each member's own machine through the Bridge, or in the vendor's own cloud on the member's plan. Cookbook hosts the board and the memory, not the agents.
Who pays for inference	Your agents' own model calls. Zep meters credits per plan with overage.	Each member's existing subscription. No API keys, no credits resold.
Memory	The whole product: temporal graphs from many sources, automatic invalidation of outdated facts, Observations on higher plans.	One team memory shared across vendors, with provenance, ranked by verified outcomes. Stays when people leave.
Receipts and attribution	Access control, retention policies and audit trails at the data layer; API logs for 1 day (Flex), 7 days (Flex Plus) or 1 year (Enterprise).	Every run returns a receipt: person, agent, model, duration, cost, verdict. Every change carries Person and Agent.
Approval model	Not applicable: a memory service, not an agent runner.	A standing policy per teammate (allow, ask, off). Nothing changes on a machine without the owner's click. Agents cannot delete.
Pricing	Free $0 (10,000 credits per month, 2 projects); Flex $125/month (50,000 credits); Flex Plus $375/month (200,000 credits); Enterprise custom.	Free to start.

Choose Zep if

You are building a production agent and need a governed memory graph with retrieval latency guarantees, SOC 2 and HIPAA options.
Your memory is about your product's end users and business data, at a scale measured in credits and requests per minute.
You want a memory service your engineers integrate through SDKs.

Choose Cookbook if

Your team is people using Claude, ChatGPT, Gemini or Kimi day to day, and the memory that matters is decisions, gotchas and conventions.
You want that memory next to the task board and the files, written and read by every vendor's agent, with a receipt for who added it.
You do not want to integrate an SDK; you want agents to connect from their own settings and start knowing what the team knows.