We are pleased to announce a significant update to our AI request routing system.

Starting with [#Beta], all AI requests will be centralized through OpenRouter, replacing the previous method of routing messages to individual AI providers. This change enables us to manage AI providers more efficiently and streamline the overall process.

OpenRouter will serve as the primary platform for routing AI requests, improving operational efficiency.

However, requests to the Mistral provider will continue to be routed through Mistral AI servers, ensuring that all data remains within the EU.

We are also excited to introduce Anthropic, featuring the Claude 4.5 model, which will be available for use in our /ai chat features.
Use /aiprovider to change your AI Provider.

Our Privacy Policy has been updated to include OpenRouter and provide detailed information on how and when data is routed to each provider or service.

🗓 As always, this update is rolling out first to users with Beta mode enabled and will gradually reach more users in the coming weeks through an A/B release process, ensuring the seamless experience.

I recently discovered a scenario where Sophie couldn’t enforce the /welcomecaptcha in some chats. This happens when Telegram doesn’t send updates to bots about new members joining.

[New Feature: Telegram’s "Approve new users" support]
To address this, we’ve added support for Telegram’s "approve new users" feature. When enabled in your chat settings, Sophie will now automatically send a captcha message to the private chat of users requesting to join. Once the user completes the captcha, Sophie will approve their request.

[Automatic Handling of Incomplete Captchas]
But that’s not all! Users who fail to complete the captcha within 48 hours will now be automatically banned or rejected from the chat, even if the "approve new users" setting isn’t enabled.

[Federation Module Modernization]
We’ve also introduced a new Federation module in the /help menu, featuring the new /finfo command. This is the first step in modernizing the existing Federations module, with more improvements on the way. I’ll be rolling out new commands and features gradually, using feature flags to ensure the best possible experience.

🗓 This update is rolling out first to users with Beta mode enabled and will gradually reach more users in the coming weeks through an A/B and feature-flags release process, ensuring a seamless experience.

[Introducing AI Filter Handlers]!

We’re excited to announce a powerful new feature: AI Filter Handlers. Powered by Mistral AI, an industry-leading AI provider known for its commitment to data privacy, this feature allows you to intelligently determine whether a filter should be triggered based on the content of any message, whether it’s text, stickers, videos, GIFs, or pictures.

[How to Use It]:
Simply use the following pattern when adding your filters:
/addfilter ai:Your prompt

[Example]:
/addfilter ai:Has anything regarding money or cryptocurrency

Of course, you can choose and combine a wide range of actions for your filters, such as:
- Sending an AI-generated reply
- Deleting the message
- Warning the user
- Banning the user

The Filter's wiki page was updated with even more information about this new feature!

🗓 As always, this update is rolling out first to users with Beta mode enabled and will gradually reach more users in the coming weeks through an A/B release process, ensuring the seamless experience.