Curated Data vs a DIY RAG stack

Retrieval finds text. Approval decides what your agent should trust.

Assembling your own vector database, embeddings pipeline, and retrieval layer gets you search. It does not get you a review workflow a business owner can run, an audit trail, or a source that only ever serves what was approved. Here is where the two approaches actually differ.

Approval gate built in, not bolted on
No vector database or retrieval pipeline to operate
A review workflow business owners can run themselves
Free tier to start

Approval is the product, not an add-on

A self-built stack retrieves whatever is textually similar to the query, including a document nobody reviewed. Curated Data serves only the current approved revision of a concept doc, enforced at one code chokepoint, so what an agent reads is never one query away from something unvetted.

Nothing to run, patch, or scale

No vector database, embeddings job, or retrieval pipeline to operate yourself. The hosted MCP server and REST API serve the same approved concepts your team writes and reviews, from day one.

A review workflow a business owner can run

Writing a concept doc looks like a shared document with an approve button, not a pull request. The people who actually know the policy or the pricing rule can own it end to end, without waiting on an engineer to update an index.

An audit trail your security review can read

Every approved revision keeps its author, reviewer, and approval time. When someone asks which version was live on a given date, that is a lookup, not an archaeology project.

Questions

Doesn't a good retrieval pipeline solve this already?: Retrieval finds the passages closest to a query. It has no concept of approved versus draft versus stale, so it can surface a document nobody signed off on just as confidently as one they did. Curated Data adds the review step and only ever serves the approved version.
We already have engineers who can build and maintain a RAG stack. Why not just do that?: You can, and some teams should. The tradeoff is total cost of ownership: someone owns the vector database, the embeddings job, the retrieval tuning, and the index freshness, indefinitely. Curated Data is hosted, and the review workflow is built for the policy owner, not the engineer maintaining the pipeline.
Can we use both?: Yes. Many teams keep a broader retrieval corpus for general search and use Curated Data as the governed lane for the high-consequence answers: pricing, policy, entitlements, anywhere a wrong answer is expensive.
What does curation cost us that a self-built index doesn't?: Writing and reviewing the concepts that matter most. That is real effort, but it is bounded: you do not have to curate everything, only the answers where a wrong guess is costly. The free tier covers up to 200 approved pages, which is enough for most teams' first pass.
What does this cost to start?: Free, for up to 200 approved pages and 5 members. Team is $99 a month for larger rollouts.

Migrating off a self-built RAG stack?

Tell us what you have built so far and we will help you figure out what moves and what stays.

Stop maintaining the pipeline. Start approving the answers.

Start free, write your first approved concept, and connect the agent you already have.