We already have a help center and a wiki. Why this?

Keep them. A wiki holds everything, including drafts and stale pages, and an AI reading it cannot tell which is which. Curated Data holds the smaller set of answers you must get right: policies, pricing, terms. Each one has a named owner and an approval step, and it is what your AI answers from.

What happens when a policy changes?

The edit becomes a draft while the current approved version keeps answering. Once the new version is signed off, every assistant you connect answers from it on the next question. Earlier versions stay in the history.

Can a customer-facing bot see internal-only guidance?

No. Every connection gets its own scoped key and only retrieves what it is cleared to read. Reps can see the internal playbook while the customer-facing bot only sees customer-safe answers.

Do we need engineers to set this up?

Not for the day-to-day. Writing and approving answers works like a shared doc with an approve button. Connecting an assistant is a one-time step for your IT team: one config block and a scoped key, with nothing to host.

Which AI tools does it work with?

Anything that speaks MCP, including Claude Desktop, Claude Code, Cursor, Windsurf, VS Code, and the OpenAI Agents SDK. There is also a REST API for tools that are not MCP native.

What happens if we switch AI vendors later?

Your knowledge layer carries over. MCP and REST are provider-neutral surfaces, so a new assistant connects with a new scoped key and reads the same approved answers — no migration, no re-approval. Underneath, every answer is a plain markdown document in an open format, not a proprietary store. The approval work your team does here outlives any one AI vendor choice.

How does it stay current after the initial import?

Two ways. Your team edits and re-approves answers as policies change — the usual path. And the assistants themselves help: the MCP connection instructs them to check the approved knowledge first and, when they hit a question it cannot answer, to file the missing fact as a suggestion. Suggestions land in your review queue as drafts. Your team approves before anything is served — agents propose, people decide.

How do we know an answer is current, and who approved it?

Every answer traces back to a specific approved version that records its author, reviewer, and approval time. Approvals and rejections land in an audit log you can hand to a security review.

For customer success, deal desk, and support teams

The answer your company approved.Not the AI's best guess.

Curated Data is where your team keeps the approved version of your policies, pricing, and terms. Every AI assistant you connect answers from it. When something changes, update it once and every answer follows.

Start free Watch it go draft to approved to live

Free tier · no credit card · no infrastructure to run

Your AI assistantanswering from approved knowledge

Asked

Can I refund an annual plan 45 days after purchase?

The approved answer

Refunds are available within 30 days of purchase. After 30 days, offer prorated account credit with manager approval.

Grounded in knowledge your team approved

source: policies/refundsapproved by CS leadership

Approved by your team
Every change tracked
Internal and customer-facing kept separate
Your knowledge stays portable

Your AI is already answering for your company. Make sure it answers correctly.

An AI without your company's knowledge fills the gaps with whatever it learned from the internet, and it does so confidently. That is how a help bot quotes a refund window you never offered, or a rep repeats last quarter's discount floor.

Curated Data gives every assistant one approved answer, written in your team's words, signed off by the owner you chose, and traceable when someone asks where it came from.

Sample

An AI on its ownno sources connected

A customer bought an annual plan 45 days ago and wants a refund. What do I tell them?

Annual plans almost always come with a 60-day refund window, so you are safe to approve this one in full.

no source · a guess

The same AI + Curated Dataconnected: curated-data (MCP)

A customer bought an annual plan 45 days ago and wants a refund. What do I tell them?

Found the approved policy.

The refund window is 30 days, so at day 45 this one misses it. Offer prorated account credit instead. That needs manager approval.

verified · policies/refunds · v7 · approved

Same question. The difference is whether the answer came from knowledge your team approved, or from a guess. Grounding will not make an AI perfect, but it shrinks the room for guessing and makes every answer traceable to a source you approved.

Built for the teams who answer for the company

You own the refund policy, the discount matrix, and the support playbook. Curated Data keeps them current and approved, so every AI you connect answers from the same source.

Heads of Customer Success

Your CSMs ask an AI before they ask you. Keep playbooks, entitlements, and renewal policies approved and current, so every answer matches the one you would give.

See how it answers

Deal Desk & RevOps

A quote built on last quarter's terms costs real money. Keep discount thresholds, payment terms, and approval rules in one approved source your sales assistants quote from.

See how it answers

Customer Service

One policy, two audiences. Reps get the internal guidance, your help bot gets the customer-safe answer, and both stay current when the policy changes.

See how it answers

Also built for:Sales Marketing IT HR Legal See all solutions →

Write it down once. Approve it. Every assistant uses it.

Three steps take an answer from a page someone on your team wrote to something every AI you connect can cite: Write, Approve, Serve.

Today

every AI on its own

Each assistant stitches together its own context: no review, no versions, no way to trace where an answer came from.

With Curated Data

one approved source

Every assistant answers from the same approved versions: tracked, scoped by key, and traceable to a source your team signed off.

What your team does in it

The four surfaces you will live in: from writing an answer to the audit trail it leaves behind.

Write it down once, in your team’s words

Every policy, definition, and product detail gets one page with a named owner, an approval status, tags, and links to related answers. This is the saved page as your team sees it.

policies/refunds.md

Approved

Ownercs-leadership

Typepolicy

Tagsbillingcustomer-facingrefunds

Refunds

Self-serve refunds are available within 30 days of purchase on every plan, including annual. After 30 days, offer prorated account credit instead; credits need manager approval before they are issued.

Quick rules

0 to 30 daysfull refund to the original payment method
31+ daysprorated account credit, manager approval
chargebacksescalate to billing · never self-serve

Linked pagesaccount creditannual planscancellations

Pages link together, so related answers stay connected
Plain, portable documents underneath. Never locked in.

The approved answer, wherever your team uses AI

Assistants connect with a scoped key and get back the current approved version, limited to what that connection is cleared to read.

key: help-center-botreads: customer-safe only

MCP · search_knowledge

→ search_knowledge("can I refund an annual plan after 45 days?")

{
  "path": "policies/refunds",
  "revision": 7,
  "status": "approved",
  "body": "Refunds are available within 30 days of purchase…"
}

# drafts and in-review versions are never returned

Hosted for you: nothing to run or scale
Customer-facing tools only see what they are cleared to share

Answers you can defend to a customer, a boss, or an auditor

Every answer your AI gives traces back to a version a named person approved: recorded, versioned, and scoped. The controls your security team will ask about are built in.

Built on the open Open Knowledge Format

A named person approved every answer
Nothing unreviewed is ever served. Each change is signed off and recorded in the audit log.
Customer-facing stays customer-safe
Each connection has its own scoped key and only sees what it is cleared to share. Internal guidance stays internal.
Every version kept
Every page records its versions, author, reviewer, and approval time.
Isolated per workspace
Every workspace is its own tenant. One company's knowledge never crosses into another.
Security review & DPA
Available on request for Enterprise. OIDC SSO is on the roadmap; SAML and SCIM are planned after.
No lock-in
Plain, portable documents underneath, built on the open Open Knowledge Format. Your knowledge stays yours.

claude_desktop_config.json

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

Works with the AI tools you already use

Claude, Cursor, or any MCP compatible assistant connects with one config block and a scoped key. There is nothing to host and nothing to install. For everything else, the same approved answers are available over a REST API. Both surfaces are provider neutral: swap models or vendors later and the same approved knowledge keeps answering.

Hosted for you: nothing to run or scale
One scoped key per assistant, revocable any time
REST API for tools that are not MCP native

Send your IT team the docs →

Works with any MCP client

Claude Desktop
Claude Code
Cursor
Windsurf
VS Code
OpenAI Agents SDK
plus REST for everything else

Before you rely on an AI for answers

The questions to ask first

Curated Data is deliberately narrower than a wiki or an enterprise search tool. These are the boundaries and controls that make it dependable.

Explore examples →Read the full docs →

Start free. Grow when your team does.

Every plan includes the full write, approve, publish workflow and works with every assistant you connect. Higher tiers lift caps and add controls.

Compare plans →

Free

$0forever

For a single team getting started.

Up to 200 concept docs
5 members
MCP + REST access

Start free

For production

Team

$99/mo

For teams putting approved answers behind AI in production.

Up to 5,000 docs
25 members
100,000 API requests / day

Start free

Enterprise

Custom

For scale, SSO, and dedicated support through sales.

OIDC SSO on the roadmap
Expanded member limits
Custom rate limits & SLAs

Talk to sales

Give your AI the approved answers

Start at whatever depth fits: watch it work, build on the free tier, or bring your security team.

Try the guided demoWatch an answer go from draft to approved to live, in the browser. No signup.Open the demo Start freeCreate a workspace, approve the answers your AI needs, and connect it. No credit card.Create your workspace Talk to salesWant someone to walk your team through it? We'll set up a short call. Security review is available by agreement; OIDC SSO is on the Enterprise roadmap.Talk to sales

Not ready to set one up yet?

Leave your work email and we will send the occasional note when we ship something that changes what your AI can safely answer.

The answer your company approved.Not the AI's best guess.

Your AI is already answering for your company. Make sure it answers correctly.

Built for the teams who answer for the company

Heads of Customer Success

Deal Desk & RevOps

Customer Service

Write it down once. Approve it. Every assistant uses it.

What your team does in it

Write it down once, in your team’s words

Someone you trust signs off before any AI uses it

The approved answer, wherever your team uses AI

Every change on the record

Answers you can defend to a customer, a boss, or an auditor

A named person approved every answer

Customer-facing stays customer-safe

Every version kept

Isolated per workspace

Security review & DPA

No lock-in

Works with the AI tools you already use

The questions to ask first

Start free. Grow when your team does.

Free

Team

Enterprise

Give your AI the approved answers

Opening Curated Data