Roadmap
Help Center
Changelog

Help me improve BoltAI

For a bug report, please include the following information:

  • Your OS version.

  • Your BoltAI app version.

  • Are you a Setapp user.

  • The AI provider & model you’re using.

  • Steps to reproduce the issue.

  • The error message

Extremely high energy usage.

When I put a discussion in the trash, a loading indicator appears on the navigation bar. However, when this indicator is active, it triggers a massive energy consumption spike and significant overheating. The issue goes back to normal when I hide the navigation bar and then make it reappear. OS Version : 26.5.1 BoltAI Version : 2.13.2 (Build 71) Are you a Setapp user : No Provider AI : Crof.AI Model : DeepSeek Pro V4 Steps to reproduce the issue : Open the application and navigate to any discussion page. Move the/a discussion to the trash. Observe that a “Loading…” indicator. Do not interact with the navigation bar — let the loading indicator remain active for several seconds. Notice the device starts overheating and the battery drains rapidly due to increased energy consumption. To temporarily resolve the visual/performance issue: Hide the navigation bar. Then make it reappear. Observe that the loading indicator disappears and the device returns to normal temperature/power behavior. You can see the video for more details.

Valentin Arnould 10 days ago

option to Archive a Project

So I'm getting more and more Project folders. When I start a new project, I create a new Project, and let's say I work on it for two weeks with 20+ conversations, files, etc. Now the project is finished, but I don't want to delete it, as I "might" need something from it in the future. A collapsible Archive folder that holds Archived Projects would be great.

David 10 days ago

Support Gemini 3.1 and 3.5 Flash for Google Vertex AI Provider

Problem Currently, the Google Vertex AI provider in BoltAI is missing recent models, such as the Gemini 3.1 series (including 3.1 Pro Preview) and Gemini 3.5 Flash. Additionally, other non-Gemini models offered through Vertex AI are not supported. Because the AI models evolves rapidly, it is difficult to maintain and update a hardcoded list every time a new model is released. Proposed Solution Add Gemini 3.1 Pro and 3.5 Flash next release for a quick fix. For the long run, add a custom "Model ID" input field within the Google Vertex AI provider settings. This would allow users to manually enter the exact ID of the language model they wish to use. Benefits This approach future-proofs the integration. Instead of waiting for Daniel to release an update with a new model, users can instantly access the latest models (like Gemini 3.1 Pro Preview and 3.5 Flash) or any other supported Vertex AI model simply by entering the corresponding model ID.

Anıl Yuvalı 12 days ago

1

in BoltAI - MCP tool names use single-underscore mcp_ prefix → Anthropic rejects Claude (OAuth) requests with HTTP 400

When a Claude/Anthropic request includes MCP tools, BoltAI names them mcp (single underscore, e.g. mcpsearch_youtube_z50gm8). Anthropic classifies any tool whose name starts with mcp (single underscore) as a third-party MCP tool and bills it to "extra usage," returning: 400 invalid_request_error: "Third-party apps now draw from your extra usage, not your plan limits…" On accounts with extra usage disabled, every MCP-tool request fails. Claude Code's own convention is a double underscore — mcp — which Anthropic accepts as first-party. Verified fix: renaming mcpsearch_youtube_z50gm8 → mcp__search_youtube_z50gm8 (only change: one extra underscore) turns the 400 into a 200. Requests without tools, or with tools that don't use the mcp prefix, already work. Request: name MCP tools with the mcp__ double-underscore convention. Tried it with https://github.com/router-for-me/CLIProxyAPI and https://github.com/automazeio/vibeproxy (and used in AI Providers - Anthropic endpoint and pointed it to CLIProxyAPI)

David 14 days ago

4

Cross-Chat References

Allow referencing or including context from previous chats in a new conversation. What I'd like: Ability to attach/link a previous chat (or multiple chats) as context to a new one Ideally searchable chat history so finding the right one is easy attach/link selectable or all history from Projects Why it matters: Work often spans multiple sessions — being able to pull in context from a past conversation without copy-pasting saves time and avoids losing important decisions/outcomes.

David 22 days ago

4

Tabbed Parallel Chats

Add a tab bar to the chat interface (at the top) so multiple conversations can run simultaneously without switching between separate windows. It would show only (active?) chats at the top… something like in the screen shoot.. Each tab runs independently (LLM calls don't block each other) Easy switching between tabs, ideally with keyboard shortcuts (cmd+left/right)

David 22 days ago

BUG: UI hangs when running mcp_add_memory_tctnzi tool

OS: macOS 26.4.1 Version: BoltAI 2.12.1 (build 68) - Setapp LLM: Anthropic Claude Sonnet 4.6 I instructed the LLM to commit something in particular to memory. It appears to have called the mem0-memory-mcp server installed. Although it’s not clear to me if that’s what Bolt tried to use, or whether it’s using the built-in Bolt memory tool. I can see in the UI what you see in attached screenshot. The conversation hang after it tried to use this memory tool. I had to manually terminate the hung prompt response. I can see that the number of memories in the Bolt Memory tool didn’t increase. So if that is the memory system it’s utilising in this chat, the hung attempt to add a memory did not get so far as actually adding a memory.

Inspired Earth about 1 month ago

2

Google Deep Research models not working

When connecting through “Google AI” AI Providers and when selecting any Deep Research model there is an error: Error This model only supports Interactions API. I have all MCPs and plugins disabled… all other Gemini models work. https://ai.google.dev/gemini-api/docs/interactions/deep-research i guess you would need to implement an additional api endpoint for this to work?

David about 1 month ago

Choice of local models for STT

Not having a choice in speech-to-text models for the dictation feature is very limiting. The two pre‑selected models are not great. Silences are always replaced with lots of random words. And the user doesn't have the ability to pick from a whole variety of excellent speech-to-text models now available. Look at something like Mac Whisper and the flows that exist there.

Thinkun about 1 month ago

4

Built a Hermes Agent gateway plugin for BoltAI — full markdown + slash commands

Hey, I've been using BoltAI v2 with Hermes Agent (https://github.com/NousResearch/hermes-agent) and hit two annoyances with the official OpenAI-compatible gateway: Markdown gets flattened — headers, tables, code blocks all come through as plain text Hermes slash commands (/help, /status, /new, /model, etc.) aren't intercepted, so they get sent to the LLM instead of running locally and therefore they dont work properly as AI agent cant work with some directly… So I built a small hermes drop-in plugin that fixes both. It's a separate gateway adapter (subclasses the official one, doesn't fork it), runs on its own port, and lives here: https://github.com/Dworf/boltai-hermes-gateway Install: hermes plugins install Dworf/boltai-hermes-gateway --enable Then in BoltAI just follow the official Hermes setup guide (https://help.boltai.com/en/help/articles/9536908-how-to-use-hermes-agent-in-boltai) but point it at port 8643 instead of 8642. Full markdown renders correctly, slash commands work as expected, streaming is preserved. Should also work with any other OpenAI-compatible client that supports markdown + streaming (TypingMind, OpenWebUI, LibreChat, etc.) — untested though (and yeah BoltAI has no competition in this space :) ) Question for the BoltAI team / community: if there's interest, I'm happy to implement a commands discovery endpoint on the plugin side so that — if BoltAI ever wants to add it on the client side — you could get interactive slash command suggestions, autocomplete, and /-menu UX natively. Let me know if that's something worth building toward. Should be easy from the plugin side… Thanks!

David about 1 month ago

2

model selection of model capability based on output (text, image, video,...)

I really like the new model selection… and love the favorites. but when working with model collections like (image generation, video generation,…) it would be good to have filter option to only show models that can do image generation or video generation,…. i would be happy to just tag each model if it supports for example image generation… i think you could pull this data as well from openrouter for example…

David about 1 month ago