Chat & Research
Attachments, the web, voice and models
Attach files
Click + beside the message box, or drop files onto the conversation, to send them with your next message: a screenshot to rebuild, a spec to follow, figures to chart.
- Up to 6 files at a time.
- Images (PNG, JPEG, GIF, WebP) up to 3.5 MB each.
- PDF, Word, Excel and CSV files up to 4.5 MB each.
- Text and code files, such as
.txt,.md,.jsonor.html.
On a Prototype tab, paste a web address into your message to have its design matched. If the page cannot be read, the reply says so; attach a screenshot instead.
Your project's documents
Every document in the project marked available to the AI is used for each message: sent whole when they fit, otherwise searched for the passages that answer you. The folder button beside + narrows that to the items you tick; with nothing ticked it reads All available, and all available documents are used. If it reads None available, no document in the project is available to the AI yet. What the AI can read
Search the web and read websites
A Chat & Research tab can search the web and read web pages. Ask about something public or current, or name a website, and the AI decides whether to look it up. A web address can be typed in full or as a plain domain, such as example.com.
- While it works, the top of the answer shows each step, such as Searching the web for … or Reading example.com, with a mark when the step is done or could not be done.
- Under the answer, the Sources badge opens the pages it read, as links, with any project documents it used. When it answered from search results without reading a page, the list is headed Search results used. Under each answer
- In one answer it can run up to 5 searches and read up to 15 pages, and up to 10 pages of any one website.
- If a page cannot be read, or web search is not available, the answer says so.
- Only public websites can be read. Text on a web page is treated as information, never as instructions to the AI.
The Web switch beside the message box turns this on or off for the tab. It is on for every new tab, and each tab keeps its own setting. With it off, the AI does not search or read the web, and says so if a question needs it. Each search, each page the web service reads, and each time it lists a website's pages to choose which ones to read (when it reads several pages of one site), uses a little credit, shown under Chat & Research in Billing. A page read directly (when web search is not set up on your Hub), and a search or page the web service could not serve, cost nothing.
Speak instead of typing
Press the microphone beside Send and speak. When you pause, your words are turned into text, put in the message box and sent. Press the microphone again before you pause to stop early: the text stays in the box for you to edit or send.
- Up to a minute at a time. If nothing is heard for about eight seconds, it stops without charging anything.
- It works in any modern browser that can use a microphone. The recording is not kept.
- If your browser has blocked the microphone, a message says so: allow it in your browser's site settings, then try again.
- Speaking uses a little credit by the length of the recording, shown under Chat & Research in Billing.
Choose a model
The menu beside the message box picks the model.
| Claude Sonnet | The default. Fast, and right for almost everything. |
|---|---|
| Claude Opus 5 | For the hardest, most detailed builds. It uses more credit. If it is not available to your account, the reply says so and builds with Sonnet. |
Signed in to VeraGen? Press Guide me on any screen for a walkthrough, or ask Odin.