Titan AI Web access
Choose which sites Titan AI can reach, and understand what that setting does and does not cover.
Titan AI can read the open web when it needs to, not just your Amazon data. Web access, on the Web Access tab in Settings, is where you choose how far that reaches.
There are three modes. Restricted is the default for new accounts.
This is a control over which sites Titan AI can reach. It does not make reading a page safe by itself, and it is not a defense against a page that tries to manipulate Titan AI while it is there. Read what a narrower list does and does not cover before you rely on it.
The three modes
| Mode | What it means |
|---|---|
| Off | Titan AI cannot reach the web at all. It works only from your Amazon data and anything you share with it directly in chat. |
| Restricted (default) | Titan AI can reach the domains on your list below, whether it fetches a page directly or opens it in a browser. Two narrow exceptions apply to both, covered under Amazon images and Package registries. |
| Full access | Your list does not apply. Titan AI can reach any site on the web. What that means for the browser is covered under Browser automation. |
Off and Restricted apply the moment you pick them. Full access is different: it turns off a protection rather than narrowing or holding one, so choosing it opens a confirmation first, naming what it turns off. Nothing changes until you confirm.
Your domain list
The list below the mode picker is what Restricted mode enforces. Add a domain by typing it and pressing Enter or Add; remove one with the X next to it. Some entries are marked Managed and cannot be removed, these are on the list by default so Titan AI can do basic work out of the box.
You can write an entry three ways:
| You write | It matches |
|---|---|
example.com | That exact site, nothing else. |
*.example.com | Its subdomains only, not example.com itself. |
**.example.com | The site itself and every subdomain. |
Most sites move you from the plain name to the www address the moment you
arrive, and an entry for the plain name does not cover that second address. So
example.com on its own can still fail to open, even though it looks like the
site you asked for. **.example.com is the entry that covers both, which is
why it is the form to reach for when you are adding a site you want to browse.
If you do hit this, Titan AI will tell you that a redirect is what stopped it,
so you know to reach for the **. form rather than guessing.
You do not need to remove the narrower entry first. Adding the wider one is enough, and it works the same way for an entry marked Managed, which you cannot remove.
The list has no effect while web access is Off, and no effect in Full access, since neither mode enforces it.
Add the sites you need Titan AI to reach. A few kinds of entry are never
allowed, such as a wildcard over a whole domain ending like .co.uk, or an
IP address. If one is refused you will be told why.
Browser automation
When web access is on, Titan AI can also open a page in a browser and take a screenshot, not only fetch it directly. It can click and type on the page too, the same way you would, not only look at it. That is a separate path from the direct fetch described above, and your mode changes what it means for that path.
On Full access, the browser is unrestricted, same as everything else in that mode. There is no list to apply, so it can open any page except your own Titan Tools pages (below).
On Restricted, the browser is held to exactly the same list as your direct fetches. The browser gets no network access of its own: everything it loads, including the resources a page pulls in and anywhere a redirect sends it, goes through a filter built from your list, so a host off your list does not load. The two exceptions, Amazon images and Package registries, are the same for the browser as they are for direct fetches.
In every mode, the browser does not open your own pages in the Titan Tools app, such as a product research page. Those pages need your Titan Tools sign-in, and Titan AI never signs in to Titan Tools for you. For a product research page, upload its selected-ASIN export to the chat instead. A report download link from Titan Tools is not one of these pages.
Amazon images
Amazon's pages show their product images from two separate Amazon image sites. On Restricted, whenever your list includes an Amazon store, those two image sites are allowed alongside it, even though they are not shown in your list. This applies to direct fetches and to the browser alike, so Titan AI can open a product image itself as well as see it on the page. Without them a real Amazon page renders with no product images at all. If your list includes no Amazon store, the exception does not apply.
Package registries
Titan AI can install a code library it needs for your task, such as a spreadsheet or document library, into your conversation's own workspace. To do that, on Restricted it can always reach three package registries, even though they are not shown in your list and cannot be removed from it:
pypi.orgfiles.pythonhosted.orgregistry.npmjs.org
These are where Python and JavaScript libraries are downloaded from. A library installed in one conversation is not visible to any other. On Off, the registries are not reachable either, so Titan AI tells you a library is unavailable instead of installing it.
What this setting does and does not cover
Restricted mode governs the pages Titan AI loads, whether it fetches them directly or opens them in a browser: opening a URL, reading a page, following a link. It does not cover everything else Titan AI can do on the web, and it is worth knowing the difference before you trust the list as a complete boundary. The Browser automation section above covers the browser path, and Amazon images and Package registries cover the two exceptions, which apply to both paths.
Web search is a separate channel. Restricted mode's list does not limit what a search can return. If Titan AI searches the web on your behalf, the results it gets back can come from any site, not only the ones on your list. Treat anything that comes back from a search the same way you would treat a result from any other search engine: read it, do not assume it was pre-filtered to your list.
That is not a bug in your setting: Off still turns off all web access, including search, and your Restricted list still does exactly what the table above says. The point is narrower than "Titan AI can only see these sites." It cannot reach outside your list, whether by fetching or by browsing, apart from the Amazon images and Package registries exceptions, but it can still encounter other sites through search results.
Restricted mode shrinks how much of the web Titan AI can be pointed at, and that is real protection. It is not a promise that everything it reads is trustworthy, and it is not a defense on its own against a page written to try to manipulate an AI reading it. Treat what Titan AI reads off the web the way you would treat an email from someone you don't know: useful, but not automatically believed.