Brave Leo AI stops responding whenever the current webpage is attached as page context in Brave Stable v1.97.56 on Windows 11.
Leo remains stuck on “Thinking” indefinitely without generating any response, regardless of the AI model selected.
After extensive troubleshooting, I was able to identify the specific condition that triggers the issue.
Steps to reproduce:
Open any webpage in Brave Stable v1.97.56.
Open the Leo AI sidebar.
Enable page context so that Leo recognizes the currently open webpage.
Enter a simple prompt such as “Hello” or ask Leo to summarize the page.
Leo remains stuck on “Thinking” indefinitely without responding.
Disable or remove the attached page context.
Enter another prompt. Leo now responds normally.
Important findings:
The problem occurs with both regular webpages and YouTube videos.
All available AI models are affected.
Leo works normally when page context is disabled.
YouTube video summarization also fails when page context is enabled.
Creating a completely new Brave profile does not resolve the issue.
Restarting Brave does not resolve the issue.
The same problem occurs on a second Windows PC running the same Stable version, connected to a completely different network.
The two PCs use Brave Sync, but the issue also occurs in a newly created browser profile.
Brave Leo on Android works normally on the same network as the first PC.
Comparison with Brave Beta:
I installed Brave Beta on the same Windows PC.
Everything works correctly in Beta, including:
Normal conversations with Leo.
Questions about the currently open webpage.
Page context handling.
YouTube video summarization.
The issue is reproducible in Brave Stable but not in Brave Beta.
Environment:
OS: Windows 11
Brave Stable: v1.97.56 (October 7, 2026)
Brave Beta: Working normally
AI models: All available models affected in Stable when page context is enabled
Network: Reproduced on two PCs using different networks
Expected behavior:
Leo should respond normally when page context is enabled, including answering questions about webpages and summarizing YouTube videos.
Actual behavior:
Leo remains stuck on “Thinking” indefinitely whenever page context is enabled. Disabling page context immediately allows normal conversations.
Conclusion / Request:
Based on these tests, the issue appears to be related specifically to page context handling in the Stable channel rather than a general AI model, network, or server outage.
Since Brave Beta works correctly under the same conditions, could the development team investigate whether a fix or implementation difference in Beta is missing from the current Stable release?
One important clarification: Unlike the behavior described in GitHub Issue #59293, Leo correctly detects the currently open webpage in my case, and the page appears as an attached link at the bottom of the Leo panel.
However, the built-in “Summarize this page” command often gets stuck on “Thinking”. Manually entered prompts work on some webpages, but YouTube video summarization consistently fails.
Therefore, page detection itself appears to be working, while processing the attached page context is unreliable.
While I do see the difference you’re pointing out, I was seeing the same behavior when I opened that issue (and the related one in Nightly), in addition to what’s in the issue.
I’m pretty sure that when those issues are fully fixed/merged that the issue you’re seeing will go away as well. I’ll ping some of the devs working on that and make sure.
Same issue on Void Linux using the same version as the OP.
On a side note, the new behavior of Leo’s sidebar following me around from tab to tab is making this even worse because I don’t know if I can change my tab while it’s working, nor do I want the same Leo conversation to see all the Tabs I’m switching between while waiting for an answer. I wish we could revert to the old behavior where the conversation is tied to a specific tab and I can @ more tabs as needed manually. This “follow me around” behavior is just slowing me down and making it hard (or impossible?) to work efficiently on multiple tabs, that need difference conversations, or some don’t need and should not have a conversation open about them.
Same issue with me on macOS. The page context being included somehow stops all response and gets stuck on “Thinking…” forever. Delete the attached page context, and Leo works fine again (though obviously can’t refer to what’s on the page).
It’s not every page though. For example, summarizing worked on this page (this support thread). But 4 other tabs I tried, it failed on all of them and this is all I see no matter how long I wait:
Yeah, tying Leo to a specific tab was much more convenient. It would be great to have an option to switch between the old and new behavior without affecting separate conversations.
Made an account to say that enabling Brave AI Chat Detailed Page Content Extractionfixed the issue for me where the AI would just think only if a webpage’s content was attached. I assume this is a temp fix for the meantime.
Works on Win 11 (w/ basic Brave version), using OpenRouter (BYOM) and the pre-installed models.
Wanted to post to bring attention to this. This still occurs with 1.97.57. Whenever I attached context, it gets stuck in “thinking.” If I remove context, then the prompt processes.
No, sorry, I meant just it’s better having each tab treated as a separate entity by Leo. It doesn’t affect this particular “Thinking” issue.
Although last night I noticed it did seem to be working better. It seems there’s a fix in the works as mentioned over on Github so hopefully this will be resolved soon. I wouldn’t be surprised if it was related to work on Leo recently becoming E2E encrypted.
Same problem here. I uninstalled Microsoft edge cause the copilot sidebar stopped working,
and after investigating, i installed this brave, at first Leo Ai worked well, but lately if I have any webpage open it gets stuck thinking… Is very sad.
Leo used to work well but now gets stuck on ‘thinking’ after an initial response. I have Brave 1.97.59 (Official Build) (x86_64)on Linux Mint Mate. Before the latest update all seemed fine. I ask Leo to create Zazzle store listings for me and upload the image for reference. It works once and then gets stuck on ‘thinking’ for any subsequent query. Given up with it for now and use a different chatbot (Lumo). Would prefer to use Leo in the sidebar that another bot in a separate tab.
I’m right there with you. Since the changes, it would seem half my conversation with a 'Tools" model like Claude Sonnet results in non replies after many many attempts to receive an answer after ‘thinking’. This of course counts against whatever the seemingly arbitrary volume/duration limit is for the model. Changing models yields the same result. I found today that Claude Haiku is missing entirely as well and there are even fewer models that are ‘Tools’ capable. Sadly after just a few months in a 1 year subscription I am finding it completely unreliable to get personal work tasks accomplished and cannot trust I can even start a task since I cannot seem to get it to completion without timing out. I tried to set a Memory rule to create summaries but since it seems time agnostic, it cannot sense time spent to generate such a handoff summary in a timely manner before getting the conversation unceremoniously cut off and told I’ve hit a time limit wall with an extraordinary wait time like 22 or 17 hours… always different.
I did contact support only to be told in email I could resort to resetting my ‘SKUs’ akin to a credential reset, but when I went to do so it warned it could be a problem and reset could cause some type of failure (I can’t remember the exact message). Asking Ask Leo revealed that the resetting of SKUs was not really advisable since the timers are set and kept server-side.
So basically, there was no real viable solution provided and the models since that October 6 email have become more and more unstable and unreliable.
I had such high hopes based upon my initial interaction that gave the impression I could depend upon it much more than what actually resulted over a very short period of time. (Premium Leo subscriber since end of June, 2026).
