Feature Requests

I love the app, but it still has many bugs. I have seen that you invest a lot of time in the mobile app. I think the desktop app is significantly more useful, as many people actually work with it.

The progression from 3.7 to 4 has meant there is no longer a separate "Extended Thinking" model. Toggling Extended Thinking on or off needs to be done via a parameter in the API call to the standard Claude Sonnet 4 model on OpenRouter. Please support!

Much of the power of LLMs comes from lining up a series of separate commands, which often work far better than if they were combined into one prompt. For example: Translate this text from [language] to [language] using Model A. Edit it in [this style and tone] using Model B. Fact check using the deep search tool of Model C. Prepare a briefing note based on that edit using Model D. Or: Take this audio input and transcribe it with Model A. Then format the resulting text with Model B. Then take the resulting text and create an image with it using Model C. And so on. Right now these workflows are best done in Python. Long term, it'd be amazing - and empower many others - to be able to put them together in Bolt!

Grok's deep web searching tools are now available via its API: https://docs.x.ai/docs/guides/live-search But it requires specific parameters. Please support!

I'm not sure on the desired behaviour, and whether this is actually a bug. I would like to have a "Project" prompt then have a specific "Assistant" prompt that adds to it. For an example if Project is "Sales" then the Assistant may be "Messaging". I looked in the knowledge base for this case, and I was unable to find anything. It's also not clear whether it follows Project System Prompt or the Assistant Prompt. System Instruction says "Project" but then I have the Assistant activated in bottom toolbar with no feedback on which wins. From output, looks to be the Project Prompt.

It will be really helpful if I can upload files directly into the chat By files, I mean a variety of formats: text documents, code, PDFs, and even compressed ZIP files. I believe many of us frequently refer to certain documents during our chat sessions. With a file upload feature, we could share these documents right within the chat, reducing any unnecessary back-and-forths and enhancing our productivity. Moreover, this feature could come in handy when we're dealing with tasks like data analysis or summarizing data. Quite often, these data are compiled into ZIP files.

For now, we only have 4 choices to choose only 1 in general: all, no context, x previous, x last. We need 2 choice to make it more powerful: x previous AND x last. This will be very useful when we want to keep first x messages and only last x. Usually this scenario happened with first x messages as knowledge to carry, and still understand last x messages to be last discussion to understand to.

Ollama now supports "think" and "no think" settings for reasoning models like DeepSeek R1: https://ollama.com/blog/thinking Be great if this could be supported by Bolt! R1:7b is one of the best local models for mainstream laptops but it overthinks terribly on the thinking mode.

Currently BoltAI stores all the data in a SQLite database locally. Data synchronization with this SQLite db is quite challenging and so I was thinking about migrating to CoreData (native sync with iCloud) This can be quite time consuming as it can mess up the schema/data. Vote on this one to help me priotize this feature

As an alternative to nested folders (I already have far too many folders), it would be great if BoltAI supported tags (with a keyboard shortcut to focus the "tags field" for current conversation), and allowed to search for conversations by tag/s.