Search and manage it in plain English

This is the finished example: the same list, now operated entirely through conversation. The agent reads what you want, calls the right tool, and tells you what it did.

Ask it to find things

You: Which products under £50 are made by Vireo?

Agent: Two: the Cloud Mesh Tee (£44) and the Trail Cap (£24). — it searched products with a filter of brand eq 'Vireo' and price lt 50 and read back the matches.

The agent turns the question into a real search — the same query and filter you'd write by hand — and answers from the actual results, not a guess.

Ask it to change things

With a write-capable key it doesn't stop at reading. It can edit, tidy and reorganise:

You: The Trail Runner 2 is discontinued — remove it. And bump every Vireo price by 10%.

Agent: Removed the Trail Runner 2, and updated the four remaining Vireo items with a 10% increase. Want the new prices?

Each of those is an ordinary operation — a delete, a set of edits — run through the same tools you could call yourself. Moving records between lists ("promote the best sellers into a featured list") is the search-insert-then-delete dance from the movie lesson, which the agent does end to end from one sentence.

The point

Nothing here is a special "AI mode". The agent has the same read and write tools your own code would use over the API; plain English is just the interface. That means anything you can build against Search Stack, you can also ask for — and anything the agent does, you can reproduce in code when you need it in a pipeline.

Checkpoint. You've watched an agent create, load, search and reshape a list from plain English alone. One thing separates a demo from something you'd trust in production: what happens when a request is retried. That's the last step.

Go deeper: Searching, filtering and the operations behind the answers in the reference.