Add Reference Data

A Search function’s reference data is the collection of items it searches across. Each item — a snippet of text or an image — is a candidate result. When you query the function, it returns the items most similar in meaning to your query, ranked by similarity.

Add items in the Import tab

In the console, open your function and go to the Import tab. How you upload depends on the function’s input type:

• For an image Search function — drag a folder of images straight onto the Import tab. Every image in the folder becomes a sample.
• For a text Search function — drag a CSV where each row is one text item, and Nyckel ingests every row as a sample.
• One at a time — paste text or upload a single image directly. Useful for prototyping or for very small collections.

See what you’ve uploaded in the Gallery tab

Once items are imported, they appear in the Gallery tab. Gallery is where you see, browse, and manage the full collection of samples the function will search across. Use it to spot-check that everything came in correctly, remove items, or add more.

Each item is called a “sample” on the function. Behind the scenes, Nyckel embeds every sample so queries can be compared against them by meaning rather than by exact word match.

What makes good reference data

A Search function is only as good as the data you upload. A few principles:

• Coverage. If you want users to find an answer, that answer needs to be in the data. The function can rank what’s there; it can’t invent what isn’t.
• Granularity. Decide what an “item” is. For support content, an item might be one article, one section of an article, or one Q&A pair. Smaller items return more focused matches; larger items carry more context per result.
• Quality. Garbage in, garbage out. Items that are off-topic, duplicated, or low-quality will surface as low-quality results.
• Freshness. Stale items will keep getting returned for queries they no longer answer well. Build a workflow for retiring or updating them.

How many items?

Search functions work with collections of almost any size — from dozens of items to hundreds of thousands. Quality and coverage matter much more than raw count. A few hundred well-chosen items will usually outperform tens of thousands of noisy ones.

Next

Query Similar Content — use the Search tab (and the API) to query the function and read the ranked results.