Over the past few months, llmgirl has grown from a Telegram bot into one connected product that works both in the messenger and on the web. This release brings that work together: a complete English experience, clearer onboarding, current AI models, better sign-in, branded email and a persona builder that now creates real companions.
The website, dashboard and Telegram bot now support two languages with one account and shared settings.
Choose an existing companion or build a new persona with her own character, communication style and visual identity.
New users can try a conversation immediately, while timely prompts make it obvious what to do next.
Conversations are faster and more natural with current DeepSeek and GPT-5.6 Luna models.
We have completed the English version of the product. It is not a separate copy: it is the same llmgirl account with the same balance, conversations and personas. English pages live under /en, and the selected dashboard language is stored in the user profile.
Telegram asks for a language on the first start and provides a language switch in settings. Navigation, hints, errors, billing, model selection, personas and schedules have all been localized.
We also completed a substantial English knowledge section about AI companions. It includes full long-form articles, individual metadata, topic-specific FAQs and useful links for different goals and situations.
Sign-in, registration, password recovery and email confirmation now feel like parts of llmgirl rather than generic service forms. The Russian interface prioritizes Yandex ID, while the international experience uses Google. Telegram can be linked to the same user as well.
The region is detected automatically and can be changed manually. Existing Google connections remain intact. Confirmation and recovery emails now use the llmgirl visual style and arrive in the user’s language.
We removed the extra “get free tokens” step. A trial balance is granted automatically whether a person joins from the website, Telegram or social sign-in.
After choosing a persona, the interface now clearly invites the user to send a first message. This small change matters: the conversation is ready immediately, and the user no longer has to guess what comes next.
We have also prepared a recovery credit for Telegram accounts that may have been affected by the older starting-balance issue. It will be applied to the production database after the release, together with a clear notification.
Our conversation models have been refreshed. DeepSeek now uses its current supported chat model, while GPT-5.6 Luna on Replicate replaces the previous ChatGPT connection.
Luna offers the right balance of speed, expressive conversation and cost for this product. Users still choose a model through the familiar interface, and their history and selected persona remain synchronized between the web and Telegram.
The builder used to produce a useful description but left the result on the screen. A completed form can now be saved to the user account and turned into a working persona.
The behavioral instructions are assembled safely on the server from validated fields. This preserves the intended character while keeping clear and comfortable boundaries.
llmgirl now has a complete persona section. Visitors can compare Alicia Valentine, Ivy Rain and Nova Blaze, understand the difference between gentle support, practical focus and an energetic tone, and open the companion that fits the moment.
User-created personas will expand this gallery, making it possible to choose an interesting existing character without building one from scratch. Articles are now connected to relevant personas and help readers move naturally from learning about a scenario to trying it in a real dialogue.
Once this update is deployed, we will run comprehensive automated testing against the working product: registration, email, Telegram commands, model calls, balances, personas, schedules and Web-to-Telegram synchronization. The goal is to verify complete user journeys, not just isolated screens.
The next major chapter is bot_v2. One of its first areas will be long-term memory: we will measure how well the bot retains meaningful details, improve weak scenarios and address everything discovered during production testing.
Thank you to everyone who has already tried llmgirl. The product is growing, while its purpose remains simple: a clear, beautiful and comfortable way to talk to AI on the web or directly in Telegram.
