What's new in AgentOne 1.0.3

News
1

AgentOne v1.0.3 is here, bringing several notable improvements across performance, stability, and polish. Here’s some of the key changes.

Highlights

  • The AI Model Directory is now fully powering AgentOne. You no longer need to download a new version of AgentOne to have access to the latest AI models. You’ll get new AI models as they become available (every 24 hours) without any extra steps. You can also manually refresh the model list at any time in the Model List Settings.
  • The performance of the “Run Command” tool has been drastically improved. Previously, commands with long outputs or that updated frequently would cause the app to crash or lag. In this release, we’ve addressed those issues with better virtualization and output length limits.
  • We’ve updated how long chats are handled (introduced in v0.0.18) so that it’s now even more performant than before.
  • There are now over 15,000 built-in extensions.
  • Added new models, including GPT 5.6 Sol by OpenAI!
  • New Text-to-Speech Provider: Google Gemini! You can configure your Gemini API key to use Gemini Text-to-Speech in AgentOne. Visit the Text to Speech Settings to set it up.
  • AgentOne is now live in the Microsoft Store!

Miscellaneous Changes

  • In the future, we may change how the app is bundled. This will result in a much larger app size than before (sad, we know) but much higher performance and stability.
  • Fixed a bug in the release workflow that caused some releases to be missing the files for some OSes
  • Upgraded to the lasest AI SDK version (version 7) for better performance and support for more models and providers
  • Remove outdated code in the codebase, slightly reducing the size of the app installer
  • Fixed a bug in the UI for subagents
  • Updated internal dependencies and systems to help AgentOne run faster
  • Added shadows to scrollable elements in certain areas
  • Improved the settings UI
  • Some provider logos in settings had poor contrast. This is now fixed.
  • Fix a bug where denying a tool approval request would show the wrong UI.
  • Improved the UI for blockquotes so that not all blockquotes have quote (") marks around them.
  • You can now disable self-healing markdown in settings. Self-healing markdown is enabled by default and marginally impacts performance.
  • The “Browse the web” tool now shows each URLs status as it loads rather than just a generic “Browsing 5 URLs” message. You can also read the URLs that were browsed even if you cancel the tool.
  • You can now stop the current AI response via a keyboard shortcut.
  • We updated some of our internal debugging tools to help with troubleshooting and performance monitoring.
  • Fixed performance issues when switching between two streaming chats.
  • Reworked the “Scroll to bottom” button and removed the scroll shadows from the main chat area due to the lag they caused.
  • You can now hide the “Scroll to bottom” button in settings.
  • Lowered the default virtualization threshold from 100 to 20 messages.
  • Accordions now unmount their content by default when collapsed, improving performance.
  • Provider logos are now displayed with optimal contrast and sizing without any manual calculations, improving performance and visual consistency.
  • The “Run Command” tool output is now debounced by 100 milliseconds to avoid hammering the UI with rapid output updates. This drastically improves performance.
  • Occasionally, the “Stop” button would not be available on long running commands. This is now fixed.
  • Due to a large increase in account activity, your settings are now synced every hour rather than every minute. We’ve gotta keep our servers alive!
  • Layout shift caused by scroll margins being added after scrollbars rather than in the same frame as the scrollbar has been addressed.
  • Added internal debug logs to improve diagnostics.
  • Removed lazy loading from the release notes so they appear instantly, but added it to the internal test routes.
  • When a chat is selected, the welcome message section is not shown ever (this fixed a bug where it would sometimes appear).
  • Updated to rmcp v2 to support the latest MCP security features.
  • Editing the advanced config for an extension now is a “draft”, you must click “Save” to apply changes or “Cancel” to discard.

There are more changed not mentioned above, powered by feedback from our awesome users. Thank you for using AgentOne! :heart: