Curated, verified content
Every item is checked before it ships: structure, formatting, answers that compute, explanations that match. Problems are flagged, not silently published.
OceanKB LLC · Cupertino, California
Knowledge bases built for AI.
We build state-of-the-art knowledge bases for AI solutions — curated, verified and structured, so the AI agents and apps on top of them find the right answer, cite it, and keep it up to date.
What we build
A knowledge base for AI isn't a pile of documents. It's content shaped so machines can use it safely — and so the people responsible for it stay in control.
Structured for retrieval
Knowledge modeled as small, typed items — skills, questions, lessons, terms — with stable keys, so an AI can fetch exactly the piece it needs and nothing else.
Agent access built in
An MCP server and a REST API over the same operations: agents can search, read, draft and check content with scoped, revocable access.
Humans approve, agents assist
Agents write drafts; people review a clear diff and publish. Every change is versioned, so any release can be compared or restored.
Protected by design
Entitlement checks before anything is served, per-customer watermarking, encryption at rest on devices, and least-privilege access for every key and author.
Delivered where it's used
Versioned releases reach apps and agents within a minute, and keep working offline on the device once downloaded.
How it works
- 1
Model the knowledge
We define the item types, fields and rules for your domain — what a correct item looks like, and what must never ship.
- 2
Author with AI, in the loop
Your experts and their AI assistants draft content through the connector. Checks run on every edit and explain exactly what to fix.
- 3
Review and publish
An editor reviews what changed, sends it back with notes or publishes a new version. Nothing reaches users without that step.
- 4
Serve apps and agents
Published versions flow to your apps and to AI agents through the API — with access control, audit trail and rollback.
Ready for AI agents
Our knowledge bases speak the Model Context Protocol. Add the connector to Claude or another MCP client, sign in, and the agent can work with the content directly — within the permissions you grant.
- OAuth sign-in per author; keys with read or draft scope
- Tools to search, read, draft, validate and preview
- Built-in guidance, so agents follow your content rules
- Everything an agent does is a draft until a person approves it
# An AI agent working in a knowledge base (MCP)
search_items q="percent of a number"
get_item math › bank.nodes › fd.percent-of
put_item + 3 new questions (draft)
find_problems ✓ 0 errors · 0 checks
preview_link → shown to the author as students see it
submit → waiting for an editor's reviewCase study
CheetahStudy
Our own product, and our knowledge base platform in production: a self-paced math and study app for middle-schoolers, including prep for the Math 7/8 Algebra placement test. Every lesson, question and hint lives in an OceanKB knowledge base: drafts come from people and AI agents like Claude, are checked automatically, approved by an editor, and delivered to families' devices — offline-capable and protected per family.
- 180
- math skills, each with leveled practice
- 4,500+
- questions with hints and worked solutions
- 162
- step-by-step lessons
- 23
- tools for AI agents over MCP
About OceanKB
OceanKB LLC, based in Cupertino, California, builds knowledge infrastructure for AI: the layer between what an organization knows and the AI that has to use it. We care about content that is correct before it is clever, people staying in charge of what their AI says, and systems that are simple to run.
Building an AI solution on your own knowledge?
Tell us about your content and what your AI needs to do with it.
hello@theoceankb.com