Finding Similar Records

Overview

Some screens across TotalApp — particularly in the Patent & R&D Operations and Customs & Global Trade Operations add-ons — need to find the records most relevant to what you're currently working on before they can give you a useful AI-powered analysis. Rather than asking the AI to reason over your entire list of records at once (slow, and easy to overwhelm with irrelevant material), these screens rank your records by relevance first, then hand only the closest matches to the AI.

This ranking step is called Finding Similar Records, and it works the same way everywhere it appears: your search text and every candidate record are converted into numeric representations of their meaning (not just their exact wording), and the candidates are sorted by how closely their meaning matches yours. This is the same underlying technology behind Semantic Search in My Knowledge and the Writer Tools — it's the same idea, applied to a screen's own records instead of your personal knowledge library.

How It Works

On a screen that supports this, you'll see a button like "Search Similar Rules" or "Find Similar Patents" above or inside the AI analysis form. Clicking it ranks your existing records against whatever you've typed so far, and shows the closest matches — usually the top handful, not the whole list — each with a small relevance percentage. You can then narrow that selection further (uncheck anything that isn't actually relevant) before running the AI analysis itself, which only considers the records you've kept selected.

Where It Runs

Finding Similar Records always runs locally, through Ollama on your own computer — it never proxies through TotalApp's servers, regardless of which AI engine you've selected for the deeper analysis that follows it. This means:

• Your records and search text never leave your machine for the purpose of ranking.
• You need Ollama installed and running, with an embedding model selected in Settings → Agentic — see Embedding Model for setup.
• The deeper AI analysis step that follows (interpreting the matches, producing a risk score, a diagnostic summary, etc.) is separate and follows your usual Settings → Agentic AI engine choice — it can run locally or route through TotalApp's server, independently of the ranking step above.

Graceful Fallback

If your local AI model for ranking isn't set up, or isn't reachable, screens never fail outright:

• If a simple keyword match can still find something relevant in your records, you'll see those results instead, clearly marked as a less precise fallback.
• If nothing relevant can be found at all, you'll see a message explaining that no similar records were found, with a pointer to set up an embedding model for better results.
• Either way, you can still run the deeper AI analysis directly, without a ranked shortlist — it will simply consider a broader set of records instead of a narrowed one.

For Developers

This page is the plain-language explanation. For the technical reference — the shared engine every screen's ranking is built on, its caching and fallback logic, and how a new screen wires into it — see Knowledge Engine under Engines.

Where You'll See This Today

This same underlying technology is designed to be reused by any future screen that needs to shortlist its own records before an AI analysis — the list above will grow over time.

Frequently Asked Questions