Changelog
v1.27.0
The Document Analysis feature just got improved.
- New: Claude's native PDF capabilities
- New: BoltAI supports OCR now. Right-click on a PDF and select "Reprocess with OCR"
- New: Choose where to attach documents in the chat (attach on individual messages or to the System Prompt)
- Fixed: Fixed the issue where BoltAI would not open a new chat when the sidebar is collapsed
Claude's native PDF capabilities
If you're using Claude Sonnet 3.5, you can use the new native PDF capabilities in BoltAI. Find the option in the right sidebar (screenshot below).
Note that when native PDF capabilities option is enabled, Prompt Caching will be temporarily disabled.
OCR support
For PDFs with images and tables, you can "reprocess" the document using OCR for better accuracy. Currently, BoltAI only support OCR for PDF documents.
More fine-grain settings for Document Analysis feature
You can choose to attach document content to individual messages or to the System Prompt.
- When attaching to the System Prompt, you can also choose to apply Prompt Caching automatically.
- When attaching to each individual message, Prompt Caching is temporarily disabled.
v1.26.0
- New: Gemini models support native Google Search Tool.
- Advanced Voice Mode supports more voices and with caching
- Custom keyboard shortcuts for Voice Input & Advanced Voice Mode
- Improved: Better Voice Input UI & error handling
What's new?
Grounding with Google Search
BoltAI now supports the Grounding with Google Search feature for Gemini Models. To enable it, open the plugin list and find the "Google Search (Gemini Only)".
Note that it might be more expensive than using the current Web Search plugins provided by BoltAI ($35 per 1K grounding request).
BoltAI doesn't support mode and dynamic_threshold settings in this release yet.
New realtime voices
Great news. You can now more realtime voices. It's also cheaper now as OpenAI applied prompt caching automatically.
Enjoy!
Better Voice Input UI
I've made a few of improvements to the Voice Input UI:
- Support custom keyboard shortcut
- Added the ability to cancel the Voice Input session
- Retry if the transcription fails
And that's it for now 👋
Happy belated halloween 🎃
v1.25.3
- New AI Command Behavior: Open in a new temporary window
- New: New chat window is floating by default
- New: Open chat messages in browser
What's New?
New Behavior: Open in a new temporary window
When using an AI Command, you can choose to open it in a new temporary window. No need to clean up inline chats aftereward.
You can bulk edit the command's behavior to this one.
Floating Chat Windows
A single chat window is floating by default. You can access it in any Space.
Open chat messages in browser
If you have a custom markdown viewer, you can use this to view the message in your favorite web browser.
- Thanks Adam, AR, Hans, Anıl and others for the feature suggestion.
v1.25.2
- Improved: Setting spell check & auto-correct options in the chat input field
- Fixed: Inline Whisper not working for some users
v1.25.0 (Advanced Voice Mode)
Advanced Voice Mode is here.
- New: Advanced Voice Mode using OpenAI's Realtime API.
- New: New UI for the Voice Input
- New: Read Aloud streams and plays audio in real time
What's new?
Advanced Voice mode
The long awaited Advanced Voice Mode (AVM) is released. You can now have a voice conversation with GPT-4o via Realtime API.
BoltAI supports both OpenAI and Azure OpenAI Service.
New UI for Voice Input
Voice Input UI just got an upgrade. Much better than the previous version, right?
Better Read Aloud
The Read Aloud feature now streams and plays audio in real time. It's much faster and more robust.
And that's it for now. Have a wonderful weekend 👋
v1.24.0 (build 113)
This release focuses on improving the voice input & Inline Whisper experience.
- New: Use Whisper from Groq with the new model "whisper-large-v3-turbo". Go to Settings > Advanced > Voice Settings to configure.
- New: Added suppport for custom Whisper prompt
- New: Whisper AI plugin now supports custom host & Organization ID
- New (Inline Whisper): Added the option to copy transcription to clipboard
- Improved: You can configure the keyboard shortcut for toggling the chat configuration pane (Toggle Inspector)
v1.23.0 (build 112)
- New: The o1 series models now support Document Analysis
- New: Reworked the chat input field. Fixed layout issues when pasting long prompts.
- New: Estimate token & cost for Gemini models
- Improved: Keyboard shortcut for toggling chat configuration pane (Toggle Inspector:
cmd+I) - Fixed: Cannot set up custom OpenAI-compatible server
- Fixed: Cannot use emojis in chat titles
- Fixed: Fixed an issue where the app could hang when loading user's profile picture
What's new?
This release adds some great improvement.
You can use o1 series models for Document Analysis now.
The OpenAI's new o1 model doesn't support custom System Instruction. In this release, I put your custom system instruction and document content into the user prompt.
The chat input field got a huge upgrade. I switched to AppKit's NSTextView and this fixes all the layout issues when pasting long prompts. It's also a lot faster too!
Estimate costs for Gemini models. Gemini flash 8B is my new favorite small model. It's cheap and is quite capable for simple tasks such as summarization or grammar fixes...
In this release, I've added token & cost estimation for Gemini models.
New BoltAI Workspace.
This is not related to mac app, but I've been working on the new BoltAI Workspace. It's the cloud version of BoltAI with Cloud Sync & other collaboration features.
It's not ready yet but you can join the waitlist here: https://app.youform.com/forms/kyxy1r54
Have a nice weekend. See you in the next update 👋
v1.22.0 (build 109)
- New: AI Command supports new variable
{clipboard}, use content from your clipboard - New: Move chat configuration to the right pane
- New: Custom prompt for generating chat title
- New: Added support for
*.tplfile - Improved: added cost estimates for o1 models
- Improved: filter only
chatmodels for TogetherAI - Fixed: cannot use o1 models from GitHub due to
streamparameter - Fixed: search doesn’t highlight content within code blocks
- Fixed: activate the chat input when typing
/within a different textview - Fixed: cannot rename archived chats
- Fixed: cannot clone / move inline chats
- Fixed: Anthropic’s
tool_useerror when disabling plugins - Fixed: quotes and other symbols in chat titles
What's New?
This release adds many quality-of-life (QoL) improvements.
1. Move chat configuration to the right pane
The current Popover has an UX issue: it's hard to test the chat paramters. Going back-and-forth between the popover and the chat list is annoying.
I listened to your feedback and move it to the right pane (Inspector Pane). By default, it's collapsed so it won't get in your way.
2. New variable in AI Command
BoltAI now supports a new variable: {clipboard}. With this, you can choose to prompt on your clipboard content instead of the highlighted text (variable {input}).
Protip: if you don't use the placeholder {input}, BoltAI would ignore the highlighted text.
3. Custom prompt for generating chat title
You can customize the prompt for generating chat title now. Go to Settings > General, click the gear button next to the option.
4. Other bug fixes and improvements
I also improved the overall app in this release, especially bugs related to the new macOS Sequoia.
Let me know if it works well for you.
P.S: I'm still working hard on the new iOS app and Cloud Sync. Sign up for early access here: https://app.youform.com/forms/yplbdvx7
Annnnnd that's it for now 👋
v1.21.0 "New Command"
- New: AI Command now supports AI Plugins (only works with the “Open in new window” behavior for now)
- New: A new inline implementation: better compatibility & support cancellation.
- New: Customize how you want the AI Command & AI Inline to work within an app. Go to Settings > Advanced > App Specific Settings.
- Improved: Improve chat list performance (pagination)
- Improved: Added support for custom keyboard layout like Dvorak
- Fixed: OpenRouter integration fails with o1 models
- Fixed: Main Window shows twice on macOS 15
- Fixed: Error related to non-whitespace text with Claude
What's New:
The most notable change in this release is the new implementation for the AI Command feature.
- AI Command now support AI Plugins. Here is a demo video where I use the Instant Command on a video in Finder:
-
New implementation. Previously, BoltAI uses the Accessibility C APIs to get the highlighted text on the active app. In this release, I've switched to a more modern APIs and it should be more robust.
-
More advanced settings. In this release, you can choose how BoltAI reads the highlighted text. Go to Settings > Advanced > App Specific Settings to see the default settings. If you're having troubles with an app, try to use a different strategy.
- Accessibility: BoltAI would use the Accessibility API to fetch the highlighted text. This works best for native apps such as Apple Notes, Xcode or TextEdit...
- Clipboard: BoltAI would emulate Copy event to get the content of the highlighted text. This works best for a web browser or web-based apps such as Discord, Slack, Notion...
- Smart (default): BoltAI checks if the app is web-based or native and would use the proper strategy.
- Other improvements:
- BoltAI supports custom input keyboard layout like Dvorak
- You can cancel the current AI Command or AI Inline request
I've added more than 30 features in the last 3 month. With this release, I'm going to increase the price soon. If you're still on the fence, now is the best time to lock in the current price.
That's it for now 👋
v1.20.5 (build 106)
- New: new import & export UI with more logging
- New: Added support for OpenAI's latest models:
o1-preview&o1-mini - Fix: fixed a bug with Data Import & Export
- Fix: fixed an issue with iCloud sync for default plugins
What's new?
This release improves the Data Import & Export feature. You can see the detail logs on the UI and on the Console app.
BoltAI also supports exporting Memory in this version.
