What Oda can do in the browser

Oda works in the browser you share with it in a chat: it opens tabs, reads pages, and clicks and types on them while you watch and can take over. This page lists what Oda can do there, the limits of each operation, and what its reports mean; for the controls you share with Oda, see The browser in Oda, and for running the Oda app itself in a browser window, see Oda in the browser.

Open and manage tabs

Oda lists, opens and closes the browser tabs of a chat. A tab belongs to one chat, and Oda works in it only while that chat is on screen.

• The list runs most recent first and marks a page that is still loading, or a tab you have taken over.
• If a chat has no tab open, Oda reports No browser tabs are open in this chat.
• The browser is available in the DataMind Oda desktop app and the VS Code extension.

Load an address

Oda loads an address in a tab, or goes back, forward or reloads.

• An address without a scheme gets https://, and an address like localhost:5173 is treated as a host with a port, not as a scheme.
• Addresses that are not HTTP or HTTPS, and addresses that carry a username or password, are refused: Pass a complete http:// or https:// address without a username or password.
• A new address passes the same site checks as any page, and the same permission rules apply; see The browser in Oda.

Read a page

Oda reads a page in two ways, and can use both in one task:

• as an accessibility outline of the page's headings, text and controls, where each control gets a reference such as e1 or e2;
• as readable text, with headings marked.

When a tab moves to another site before an action runs, Oda reads the page again: The tab moved to another site before the action ran. Take a new browser_snapshot.

A very long page is cut at a size limit: The page content was cut at the size limit; scroll or narrow the task to read the rest.

Click and type on a page

• A click uses an element reference or a pair of coordinates, never both: Pass either ref, or both x and y.
• Typing needs an element reference. Oda focuses the field and re-reads which element has focus; if the focus moved, it stops: focus moved to another field. If the element is not a text field, it stops: that element is not a text field; click it, or find the input inside it with browser_snapshot.
• By default Oda replaces what the field holds before typing, and presses Enter after typing only when the task asks for it.

Press keys and scroll

• Oda presses one key from a fixed set.
• Scrolling moves the page in steps; the amount runs from 1 to 20: amount must be an integer from 1 to 20.

Wait for text to appear

Oda waits for text to appear on the page: ten seconds by default, up to the wait maximum set for the app. The result is either The awaited text is on the page. or The awaited text did not appear before the timeout.

Take a screenshot

Oda saves a screenshot as one JPEG image named browser-screenshot.jpg in the chat. It takes one only when the outline cannot describe what you need: Use it only when the snapshot outline cannot describe what you need, such as charts or images.

Some apps do not provide image storage, and there Oda cannot take a screenshot: Screenshots need image storage, which this app does not provide. Use browser_snapshot instead.

References in Oda's reports

Oda names what it found on a page with short references:

An older element reference is refused: That element reference is not in the latest snapshot. Take a new browser_snapshot and use its references. When a pick handle expires, the message names the element it pointed to.

What Oda never types

Oda refuses to type into password, payment, bank-account and one-time-code fields in every mode, before it focuses the field. It recognises these fields by their type (a password box), by autocomplete hints, and by words in their labels and attributes in several languages. The refusal reads: That field takes a password, payment detail, or one-time code. Never type these: ask the user to fill it in themselves, then continue.

Type these fields yourself in the panel, then let Oda continue.

Questions, answered

Can Oda read a page without clicking anything?

Yes. A read is its own step: Oda takes the page's outline or its readable text without clicking on the page.

What does "not in the latest snapshot" mean?

The page changed after Oda last read it, so the reference Oda held no longer matches. Oda reads the page again and continues with a fresh reference.

Why did Oda stop typing and say the focus moved?

Oda types only into the field it was given. Just before typing it re-checks which element has focus; if the focus moved, for example because the page changed or you clicked somewhere else, it stops instead of typing into the wrong field. Oda reads the page again before it continues.

How long does Oda wait for text on a page?

Ten seconds by default, up to the wait maximum set for the app. If the text does not appear in time, Oda reports the timeout.

Where does a screenshot go?

Into the chat, as an image named browser-screenshot.jpg. You can preview it like any other image in the chat; see Chat with Oda.

Why can Oda not take a screenshot here?

The app does not provide image storage, so screenshots are unavailable there. Oda can read the page as an outline or as text instead.

Which fields does Oda treat as sensitive?

Password boxes; fields that autocomplete hints mark as passwords, one-time codes or card fields; and fields whose labels or attributes name a password, a PIN, a CVV or an IBAN in several languages. Oda refuses to type into them in every mode, before focusing.

How does Oda click something the outline cannot describe?

It clicks by coordinates. A click uses either an element reference or a pair of coordinates, never both.

What does "cut at the size limit" mean?

The page was too large to read in one go, so Oda saw it only up to the size limit. Ask Oda to scroll further or to narrow the task to read the rest.