Bug report

For past few days i am noticing there's quite difference in the quality of merlin and the original ai tools have chatgpt subscription also even with same Prompt the output is very different, and i am selecting the same model

Any time I come back to continue a conversation, there is a long lag before I can even input my prompt/response. It has become worse in the last couple of days and the page now often becomes completely unresponsive for a long period, like it's about to crash.

There seems to be a bug in Projects—when I remove a source, it disappears from the interface as expected, but the model still appears to use that deleted source when answering questions. This makes it really hard to control what context the model is relying on, especially when I want to update or refine the dataset. Would be great if this could be fixed so that once a source is removed, it's truly excluded from the project context.

It is annoying that almost anytime I need to use merlin for a quick thing i need to pass the human check all the freaking time, ended up using chatgpt again because of this friction

(LLM Generated) Issue Summary: Document Access Failure in Chat Interface Timeline & Issue Description User attempted to reference tech stack document that was supposedly attached to project rules Assistant could only see filename reference but not content User explicitly uploaded document during conversation Assistant still unable to access document content Multiple attempts to share document failed Conversation degraded due to communication breakdown Technical Failure Points Document referenced in project rules not properly injected into conversation context Direct document upload during conversation not reaching Assistant No error message or feedback to user about upload failure System continuing to reference document without access capability Impact Unable to provide requested technical solution Wasted user time and effort Created significant user frustration Prevented completion of important business task Required user to consider alternative LLM solutions System Information Assistant: Merlin Interface: Chat-based Document Type Attempted: .txt file Document Name: "606 distribution single source of truth plan.txt" User Experience Impact Multiple failed attempts at document sharing No system feedback about upload status Misleading project rules suggesting document access capability Degraded user confidence in system Utterly useless and wasted an hour of my time.

Add any shared Merlin chats here, where Deep Research didn't work as expected. I'll reply to your cases and pass them on to the team if something's wrong. Thanks!

When I attach an image and ask it to analyze, it actually says that "I'm unable to directly analyze the contents of the attached image, but I can help provide guidance based on a description or details you observe." But If I upload it directly on Claude.ai , it can read and analyze the image

I tried comparing deepseek r1 to Merlin r1 the difference is quite vast. Both in thinking and in response. Is it possible to not limit Deepseek to pro users? Also a bug where the thinking response only shows in website, it's completely empty at the mobile app of merlin. It also doesn't have proper line break like the R1. Sometimes I want to read it but I can't in merlin due to the text format. It's barely readable. Likewise to it's Output being too close together in text. The train of thought sometimes leak at the main output of deepseek. When the prompt is long, deepseek barely understand the roleplay prompt, in comparison to the original deepseek, It's kind of gibberish.

Just in the last couple of days, instead of generating in-line with my prompt on desktop, the document generated looks to be in a separate document that you have to click on and then it appears in the righthand panel. It constantly causes the app to become unresponsive and slows everything down.