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. Relevance AI describes itself as specialist AI agents for every task: an agent-building platform where enterprise teams in sales, customer success, marketing, HR, operations and support build and deploy agents, with pre-built agents for research, meeting prep, prospecting and deal review. It routes across Claude, GPT, Gemini, Kimi and GLM, integrates with 1,000+ applications, and includes evaluations, tracing and job queues.
Sources for Relevance AI, read 2026-09-03: relevanceai.com, relevanceai.com/pricing. Their pricing and features change; check their pages before deciding.
| Relevance AI | Cookbook | |
|---|---|---|
| What it is | An enterprise agent-building platform: no-code builder, pre-built agents, evaluations, tracing and integrations. | 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 | Enterprise teams in sales, customer success, marketing, HR, operations and support. | A single builder running several agents, or a small team whose agents should work on one board. |
| Vendors supported | Claude, GPT, Gemini, Kimi and GLM, chosen by its LLM routing inside the platform. | Claude, ChatGPT and Codex, Gemini, Kimi, OpenClaw. Any of them can hand work to any other. |
| Where agents run | Relevance AI's cloud. | 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 | Metered through Relevance AI; its Enterprise plan lists custom vendor credits. | Each member's existing subscription. No API keys, no credits resold. |
| Memory | Not stated on the pages we read. | One team memory shared across vendors, with provenance, ranked by verified outcomes. Stays when people leave. |
| Receipts and attribution | Tracing, agent evaluations and audit logs; the homepage reports an average cost per task. | Every run returns a receipt: person, agent, model, duration, cost, verdict. Every change carries Person and Agent. |
| Approval model | Not stated on the pages we read. | A standing policy per teammate (allow, ask, off). Nothing changes on a machine without the owner's click. Agents cannot delete. |
| Pricing | Enterprise plan via sales. We could not read other tiers from their pricing page on the date checked. | Free to start. |
Cookbook (cookbook.team) is neutral by construction: it runs Claude, ChatGPT and Codex, Gemini, Kimi and OpenClaw as equals on one board, on each member's own subscription. Relevance AI also routes across several vendors, but inside its own hosted platform and metering.
Put the agents on one board with a receipt per run and a standing policy per person (allow, ask, off). That is what Cookbook is for: agents claim tasks, do them as the person who authorized them, and every change to a machine waits for that person's click. Relevance AI approaches the same goal by building specialist agents per business function.
No. Cookbook runs the agents you already have. Templates give a workspace a starting shape (files, crew, conventions), but the agents are yours.