Today we’re launching Curated Data: a place to turn what your company actually knows into knowledge your AI agents can trust. Not a scraped wiki or a nightly export, but a reviewed, versioned base of approved answers your agents retrieve when they answer.

Teams are wiring agents into support, internal operations, and sales faster than they can feed them reliable context. The models are good. The missing piece is a trustworthy source of your company’s answers, one with a human in the loop, a clear record of who approved what, and revision history for every document your agents can retrieve. That’s what we built.

The whole product is three steps: write, approve, serve. Here’s the tour.

Write

You capture what your company knows as concept docs: one clear document per thing, written in your own terms. A metric definition, a table’s meaning, a policy, the way a workflow really runs. You write in a normal markdown editor with a frontmatter form; Curated Data handles the structure and validates it as you go. Have existing markdown lying around? Import it in bulk and start from there instead of a blank page.

Docs connect to each other with simple [[links]], which build a knowledge graph of how your knowledge relates, so “active user” can point at “session” and an agent can follow the real relationship rather than inventing one.

Approve

This is the step most tools skip, and it’s the one that makes the difference. Every change stays a draft until a reviewer looks at it and approves it. The approval is recorded: who signed off, and when. And nothing unapproved is ever served to an agent. A draft someone is still working on can’t leak into a customer answer.

Approval is publication

When a reviewer approves a change, that document's revision becomes current immediately. There is no second button that republishes the entire workspace and no unrelated bundle version to manage. Drafts remain private, while each document keeps its revision and approval history.

Serve

Approved knowledge reaches your agents two ways from the same source of truth: a hosted MCP server for any MCP-capable agent, and a versioned REST API for everything else. You mint a scoped token, drop in one config block, and your agent is retrieving approved answers:

{
  "mcpServers": {
    "curated-data": {
      "url": "https://curateddata.megacorp.company/api/curated-data/mcp",
      "headers": { "Authorization": "Bearer ckd_live_••••••••" }
    }
  }
}

That token serves exactly one workspace’s approved documents, supports rotation, and is scoped by classification, so an agent only ever reads the knowledge you’ve cleared it to see. From there, the agent can search your knowledge, fetch a concept and its links, or check which version it’s reading for that document, all as first-class tools.

Open underneath, so it stays yours

Every concept doc is plain markdown with a YAML header, written in the Open Knowledge Format, an open standard. That’s a deliberate choice: your knowledge stays portable and yours, with no lock-in. Curated Data adds the curation, approvals, hosting, and serving that turn those files into something your agents can rely on, but the substance underneath is a format you could walk away with tomorrow.

Start today

There’s a free tier, no credit card, that includes the full write, approve, serve workflow, MCP, and REST. Create a workspace, write your first concept, approve it, and point an agent at it. The fastest way to see the difference is to ask your agent the same question before and after: watch it go from a confident guess to an answer it can trace back to a page your team approved.