AI Language Practice That Doesn't Make You Reboot the Conversation

What it does

BabelDuck is a web-based language-practice companion that chats with you via text or voice. It wraps several LLM backends into one interface, stores all conversation data locally, and targets beginners who find oral practice intimidating rather than power users chasing model benchmarks. Each conversation thread can carry its own system prompt and preferences, so you can run a formal business scenario alongside a casual café chat.

The interesting bit

The standout mechanic is the sub-conversation: when the AI corrects your grammar or suggests a better phrase, you can open a tangent to interrogate that suggestion—ask why, request examples, argue with it—and then pop back to the main thread as if nothing happened. It is a small interaction pattern, but it directly addresses the anxiety of “ruining” a practice flow by asking a meta-question.

Key highlights

• Branch off into side discussions about AI feedback without losing the main conversation context
• Voice input and spoken responses for oral practice
• Switch between multiple LLM backends from the same interface
• All conversation data stored locally; deployed via Docker for self-hosting
• Built-in tutorials and per-conversation preference settings to lower the barrier for beginners

Verdict

Best for self-hosting language learners who want privacy and hate losing context when asking meta-questions. If you are after a generic LLM frontend, this is a well-mannered chat client with a tutorial mode.