DEVONthink 4.2.2

What's new in this version:

New (Pro):
- Perform AI-driven research with multiple sources, e.g., arXiv, PubMed, even your databases, with the new `perform chat research` AppleScript command. Control where AI searches with the `sources` parameter and the number of results with `maximum results`.

Improved:
- The appearance of placeholder text in the interface, e.g., in some empty table views, is more consistent with the system settings
- We made some cosmetic improvements to the sidebar in the Help viewer
- You can now load external JavaScript for Automation (JXA) scripts with a `.js` extension in the Scripts menu, as well as use them for smart rules and batch processing
- Tags separated by commas or semicolons are better parsed and applied when set via scripting
- Updated application icon for older macOS releases and Sparkle

Improved (Pro):
- The automatically generated identifiers of custom metadata (see Data settings) now ignore characters reserved for our search syntax, e.g., `:`.
- Database searches via the Chat assistant, Chat - Query smart action, or the new AppleScript `perform chat research` command now filter irrelevant results.
- Revisions to the internal names of Anthropic's Claude models ensure future model updates won’t break existing scripts, templates, or smart rules. Also, the AI > Research Report and The Devonian Times templates now use Sonnet's revised name.
- Gemini Pro and Flash Lite are now at version 3.1 (Preview) via the native API and OpenRouter. Also, Gemini 3.1 Pro and 3.1 Flash handle reasoning better. Note that Gemini Flash Lite 3.1 (Preview) is more expensive than the former version 2.5.
- More reliable tool of chat assistant, the Chat - Query action of smart rules and batch processing, and the `get chat response for message` AppleScript command for evaluating mathematical expressions. In addition, the constant eps0 is now also supported.
- Ollama has improved support for the lfm2 model and GPT-OSS now reasons better.
- Ollama reasoning models (except GPT-OSS) enable thinking now only when using Settings > AI > Chat: Usage: Best (option is also available in the chat assistant). Otherwise it's disabled to speed up responses (e.g. up to twelve times faster in case of Qwen3.5).
- Nano Banana 2 replaces version 1 and now supports the Quality (HD or Standard) setting.
- The script assistant supports GPT 5.3 Codex via OpenRouter.
- We improved the AppleScript dictionary description of the `transcribe record… language` command.

Improved (Server):
- When revisiting the web sharing interface, a warning dialog will appear if you're not logged in.

Fixed:
- Repairing a database with orphaned OPML documents incorrectly converted them to RSS feeds or outlines. Fixed.
- The View > Hide/Show toolbar command wasn't disabled in the Help viewer. Fixed.
- Converting an empty rich text document to plain text or Markdown failed. Fixed.

Fixed (Pro):
- Importing OPML email attachments always converted them to RSS feeds or outlines and could create orphaned files. Fixed.

Removed (Pro):
- Removed the soon-to-be-retired Claude 3 Haiku. Bulk tasks, e.g., filtering search results, are now done with 4.5 Haiku. Note that the 4.5 model is more expensive
- Removed all Qwen3 235b models from OpenRouter, except the ones for coding. Use the latest Qwen 3.5 models instead