Ontology Studio · Beta
Build an ontology on any topic. Use it in your AI.
Name a subject, such as a 5G network, retail or coffee brewing. Draft its terms and relationships with the AI you already use, review them, then take the result into ChatGPT, Claude, Gemini, Microsoft Copilot or your own application.
An ontology is a shared map of the things in a subject, what each one means and how they relate. It works without an account and keeps your work in your browser. See where your information goes.
How it works
From a topic to a working model in three steps.
Selected stage
Start with material you can inspect
Bring selected files, notes or public pages. Keep the original passages and source details visible so a reviewer can trace where a proposed term came from.
OutputInspectable excerpts with source, location and retrieval date.
- Name the topic. Type a subject and, if you like, what an AI should be able to do with it. Pick an example to start from a finished model instead.
- Draft it with your AI. Copy the prompt Studio writes into ChatGPT, Claude, Gemini or Copilot and paste the reply back, or use your own API key. Everything drafted arrives labeled as unsourced, for you to review.
- Review and use it. Approve what you trust, add sources, grow the model, then copy the instructions and download the knowledge file for your assistant. Test it with a question you know the answer to.
For anyone with a subject
Students, specialists, hobbyists and teams: give an assistant the vocabulary of your subject so its answers use your terms the same way every time.
For builders and developers
Export the reviewed model as JSON-LD, Turtle, SKOS or Knowledge JSON, with stable term IDs your applications and graph tools can rely on.
Examples include a 5G network, retail, coffee brewing and a library catalog. Start blank for anything else.
Documentation and learning
Learn the builder before you request an account.
Step-by-step user guide
Purpose, features, a first-model walkthrough, interface terminology and worked examples for each task. Search it online or keep a downloadable copy.
Ontology primer and workshop
A 54-page teaching edition with foundations, technical reference and ten guided Studio workshop slides.
Worked exercises and safe use
Use the facilitator guide for a worksheet and worked answers. Understand local work, selected cloud saves and optional AI requests before supplying your own material.
Request Studio access
Continue the conversation with a working knowledge model.
Met Richard at an event, in a meeting or through a colleague? Tell us what you would like to build. Richard reviews requests personally and can send an invitation for an account that saves selected models across devices.
Documentation and guest mode are available without an account. An access request does not create a login or guarantee approval. If invited, you will receive a separate email to choose your own password.
Why it matters: one retail question, worked through
“What were sales by channel in the supplied sample?”
1. The data, the same in both runs
| Line | Order date | Channel | Qty | Unit price | Discount | Refund |
|---|---|---|---|---|---|---|
| L1 | 2026-09-01 | Online | 2 | 25 | 5 | 0 |
| L2 | 2026-09-01 | Online | 1 | 20 | 0 | 0 |
| L3 | 2026-09-02 | Store | 1 | 25 | 0 | 25 |
| L4 | 2026-09-03 | Online | 3 | 15 | 5 | 0 |
| L5 | 2026-09-04 | Store | 2 | 20 | 0 | 0 |
| L6 | 2026-09-05 | Store | 1 | 15 | 0 | 0 |
Six synthetic order lines, 1 to 5 September 2026. The currency is not stated.
2. The agreed meaning, added for the second run
For this project, sales means net sales: gross sales minus line discounts minus refunds. Tax and shipping are excluded. Refunds count on the original order date.
Curated example conventions. Your own project needs its own review.
3. The traceable result
195 − 10 − 25 = 160
Online 105, Store 55.
Without project definitions
- One plausible answer reports 195 as sales. That is gross sales: discounts and refunds were never subtracted.
- Another good answer asks which definition of sales to use. With no definition supplied, asking is the right move.
With project definitions
- Net sales are 160: Online 105, Store 55.
- It shows the formula, says tax and shipping are not in the file, and names the definition it used and its status.
Illustrative responses using synthetic data, not measured results. Run the comparison in your own assistant to see yours.
Use it in your own account
Take the model into the assistant you already use.
In the Studio, open Use it in your AI and pick your assistant. Each one gets the same two core files: project-instructions.txt for its instructions box and domain-knowledge.md for its knowledge or files. The full export adds two optional files, query templates and a manifest, for teams that want them.
ChatGPT
Create a Project or a custom GPT. Paste the instructions, then upload the knowledge file.
Claude
Create a Project. Paste the instructions into the project instructions and add the knowledge file to the project knowledge.
Gemini
Create a Gem. Paste the instructions and upload the knowledge file.
Microsoft Copilot
Create an agent in Microsoft 365 Copilot. Paste the instructions and add the knowledge file.
API or local model
Download one system-prompt.md that combines both, and send it as the system prompt.
Developers
JSON-LD, Turtle, SKOS and Knowledge JSON, with stable term IDs.
Make the comparison fair
Give both runs the same question and the same sample file, each in a fresh conversation. Only the context pack differs. Note the assistant, model and date, then score each answer Pass, Fail or Not applicable on the retail example’s Retail comparison screen. A question back counts as a pass when the meaning or data is missing, and an invented figure fails the answer however good the rest is.
The pack frames the conversation with your definitions; it does not change the model. Definitions are not current facts, so numeric questions always need the data in both runs. A few questions show a direction, not a statistic.
Why it works
The same principles as the whitepapers, at personal scale.
Meaning before the call
Give the assistant definitions, grain and status before it answers, instead of hoping it infers them.
Meaning needs an owner
A definition counts when the accountable person approves it. Drafts stay labeled as drafts.
Meaning is not authority
An ontology helps choose the right answer. It never decides who may see the data.
The same ideas appear in 25 Years of Lessons Learned: meaning needs a business authority (lesson 7), an identifier does not settle meaning (lesson 8), and no average offsets a critical failure (lesson 23).
Beta: generated terms and user-recorded approvals need review. The Studio supports practical business vocabularies; it does not provide formal OWL reasoning, enterprise approval authority or live connectors to whole knowledge vaults. Read the user guide for worked examples, review steps and export limits.
Your knowledge, your choices
Build from a collection you can inspect.
Upload and curate locally
Use the file upload button for PDF, Word (.docx), text, Markdown, CSV or HTML. No browser setting change is needed for ordinary uploads. Files are read in your browser; scanned PDFs may need text recognition before import. Review extracted passages, source details and draft terms before accepting them.
Add selected public pages
Supply a public page URL. The Studio retrieves it through an RJL-hosted service so you can review the extracted text. This sends the requested URL to the service and contacts the source website. It is selected-page retrieval, rather than a search of the whole web or an authenticated internal vault.
Draft with your own AI or API key
Copying a prompt into your own assistant sends nothing from the Studio. Optional OpenAI or Claude drafting uses your own funded API key: before a chargeable request you confirm the destination, and the topic you typed, or the evidence you selected, passes with the key through the RJL-hosted request service to your chosen provider. The application does not save the key. Personal chat subscriptions do not fund API requests. This provider path remains in Beta while live acceptance testing is completed.
Keep a backup; choose cloud saves
Collections, drafts and practice work stay local and belong in a full backup. Invited accounts can save selected models under the Studio cloud terms, with no automatic 90-day account expiry. Accepted private business-meaning excerpts can become part of a saved model. Review them before saving or exporting; read-only hosted shares filter private evidence and still have their own expiry.
Use only material you are entitled to process. Do not include personal data, customer records, credentials or regulated data. External providers apply their own terms and retention policies. The privacy notice explains the local, hosted and provider paths.
Next step
Need help agreeing definitions across a team?
A free sixty-minute working session on the metrics and terms your teams interpret differently. You leave with a one-page summary and three next questions.