Fixes

Syncfusion and other masked date inputs (e.g. dd-mm-aaaa) now fill correctly. Previously the month segment could be dropped because the component only accepts per-character digit input.
React/Vue/Angular controlled inputs that ignore raw element.value = ... assignment now fill via the native prototype value setter.
Native <select> option matching is more forgiving: numeric equivalence ("4" matches "04"), prefix match either direction ("Apr" <-> "April"), and substring fallback.
Custom ARIA comboboxes (MUI, Ant Design, Chakra, etc.) are detected and filled by opening the popup and clicking the matching option. Date pickers are excluded so they get typed into instead.


Improved filling logic: priority on visual analysis of form.

Switched to Opus 4.7 as recommended model.

New form-filling engine.
FormFill no longer depends on the external OmniParser service. It now uses a single call to a modern vision-capable AI model (Claude Opus 4.6 by default) that looks at the page the same way you do, combined with a fast rule-based pre-pass for common fields like name, email, address, and phone. Result: fewer moving parts, no Replicate account needed, and noticeably better matching on complex forms.

KeePass support. Connect your .kdbx password database and fill usernames and passwords
directly from it. A dedicated "Fill User/Pass" button appears in the popup; if multiple
matching entries exist for the site, you get a picker to choose the right one.

Live fill overlays. During a vision fill, FormFill now draws color-coded boxes over each
detected field so you can see what's happening in real time:
- Gray - field detected
- Green - filled from a built-in rule (e.g. autocomplete attribute)
- Blue - filled by the AI
- Orange - no suitable value found in your profile
- Yellow pulse - field currently being filled

Boxes clear automatically once filling finishes.

Easier model selection. The LLM Config page now has one-click preset buttons for the
recommended vision models: Claude Opus 4.6, Claude Sonnet 4.5, GPT-5, and Qwen2.5-VL 72B.

Reliability fixes.
- Forms inside cross-origin iframes now fill correctly.
- Checkboxes and radio buttons are handled properly, including unchecked states.
- Dropdowns that the AI occasionally missed are now picked up by a fallback detector.
- Fixed a race condition where the popup would sometimes report "complete" before the
main page had finished.
- Miscellaneous stability improvements around dynamic forms that change as you fill them.

Settings cleanup. The old OmniParser / Replicate configuration has been removed from the
options page since it is no longer needed.

Added option to fill forms based on visual page analysis (experimental). Requires Replicate API key on top of existing LLM Openrouter API key.

Fix concurrency issues in extension.

Fixed "stop filling" sometimes not responding correctly
Faster filling

Much faster filling using less LLM credits
Added proper loading bar and fixed existing loading bar issues
Fixed various minor issues

Fixed bug in last release. Added custom prompt entry.

Completely changed layout.

Can now fill forms that require multiple profiles of different people/organizations.

Fixed some security issues.
Nicer buttons.
Fixed extension icon issues.

Added a stop button to stop cancel filling.

Fixed progress bar not working.

Fixed free input text field only accepting one line.

<3 Grok 4

Improved simulated typing for data entry into form fields.
Improved simulated page click to focus form.
Improved error log messages.

When configuring a new LLM API, and exiting the config window, the LLM was not automatically selected. This is now fixed.

Fixes bug where the test profile doesn't get loaded on first load.

Fixes a bug that caused the extension not to load properly.

Fixes (hopefully) a small bug where the default profile doesn't get loaded on first start of the extension.

You can now configure your LLM of choice. Use either a self-hosted local LLM through the Ollama app (external app) or use any third party LLM of choice through openrouter.ai

Beware: if you chose to use a third-party online LLM (unlike Ollama) your personal information will be sent to the third party LLM for processing. If you don't like this: use Ollama which runs offline with a model of your choice running on your own machine.

Added status updates during filling.