Skip to main content

Built-in tools

Depending on the model and the operator serving it, a model can search the web, read a page, and generate or edit images in the middle of a reply. These built-in tools run on the operator's node you routed to, not in your browser, with one exception noted below.

What the tools can do​

ToolWhat it does
Web searchLooks up a query on the web and feeds the results back to the model, which cites them in its answer.
Web readFetches a specific page and converts it to clean text the model can quote from.
Image generationCreates an image from a text description, inline in the reply.
Image editingModifies an image — yours or one it just made — from your instructions.
Current timeTells the model today's date and time. Available on every node that offers tools.
ZeroSignal docsLooks up these help pages to answer questions about the app. This is the exception: it runs in your browser, over a bundled copy of the docs, and sends nothing anywhere.

The model decides when to use a tool while it's answering; you don't call them directly. A single reply can use several in a row.

Turning tools on and off​

Web search, Image generation, and ZeroSignal docs each have a switch under Settings → Tools and in the composer's settings popover. The Web search switch covers web search, web read, and image search, which no operator offers today.

Where they run​

When a model searches or reads the web, the operator's node makes the request, so the site sees the operator's IP address, not yours. The operator still doesn't learn who you are: by default your request reaches its node through a relay, with no identity attached. See Privacy & security for who can read your prompt on each kind of node.

Confidential nodes can offer these tools too. A search or a page read has to reach the open internet, so on any node, including a confidential one, the query or the URL goes out. That happens only when you've left Web search on: with it off, no search tool is sent with your request at all, and whatever that node's own privacy promises (Privacy & security) hold without qualification.

Seeing tools at work​

Each tool call shows up in the reply as a compact line saying what it's doing, such as Searching for "algorand consensus" or Reading algorand.co. The sites a search finds appear beneath it as clickable chips as they arrive. A run of consecutive calls collapses into a one-line summary, such as "2 web searches · 1 page read", that expands to the full list with each call's detail. Operators on an older node version don't send the detail, so their lines read generically ("Searched the web"). Once the model starts writing its answer, sources also appear as numbered citations at the end of the message.

Images appear inline in the reply; see Chatting → Generating and editing images for what you can do with them. The Image generation switch's Images per request limit caps how many one request may produce (Settings), and each image has a per-image cost (Pricing).

Not every model has them​

Tools are a per-model, per-operator capability, so the model picker shows what's available before you send:

  • Tools — the model can use web search and web read.
  • Image tools — the model can generate or edit images inside a text reply.
  • Generation / Edit — the model is a dedicated image model.

See Models & operators for the badges, and The operator network for how operators advertise what they serve. Auto mode routes image requests to a model that can handle them.