What this does
When a customer asks a question that is already answered in your blog or docs, the AI agent can search Sanity and ground its answer in real content from your CMS — with the title and slug so the customer can read more.
You pick exactly which content types are visible to the agent. Unchecked types are completely hidden, even if the agent asks for them.
How to connect
- Go to Settings → Team → Integrations → Sanity
- Enter your Project ID (find it at sanity.io/manage)
- Enter your Dataset (usually
production) - Optionally paste a read token (only needed for private datasets — public datasets work without one)
- Click Connect
We will pull the list of distinct content types in your dataset (drafts and Sanity system types are excluded automatically) and show them with document counts.
Choose what the agent can see
After connecting, you'll see a list of every content type with a toggle. Turn on the ones the agent should be able to read — for example:
- ✅ post (blog posts)
- ✅ template
- ✅ faq
- ❌ author (probably no value for support)
- ❌ webflowComponent (internal, irrelevant)
Security
- Drafts are never exposed to the agent — only published documents
- The allow-list is enforced server-side. Even if the agent tries to query a disabled type, it will be rejected
- Queries are constructed server-side from constrained inputs — the agent never writes raw GROQ
- Read-only — there is no way for the agent to modify your CMS