Numerous studies affirm how pair activities help improve the quality of learning for second language classrooms by helping develop one’s speaking and listening skills and increasing class participation. However, non-trivial logistic issues related to pair activities make them unappealing to teachers and students alike. The advancement of multilingual LLM based chatbots in recent years provide for promising opportunities for assisting learning beyond classroom hours, emulating interpersonal conversations at the learner’s convenience. The paper outlines the architecture of a browser-based tool for supplementing learning using a conversation companion for the user, customised for our university’s French 101 and German 101 classrooms. The chatbot interface is designed to improve the user’s semantic understanding of the target language using pedagogical annotation to the conversation like translations, corrections, and pronunciations, without breaking the flow of the conversation. To keep the conversations aligned to the curriculum and to hinder the AI from generating complex responses, we have built a multi-agent system using custom prompt-engineered OpenAI bots which implements adaptive logic. The tool also aims to help users enhance and revise their vocabulary, through repeated vocabulary exposure as well as reward based grammar games. The paper further contains results from the analysis of the chatbot’s performance based on preliminary user interactions. We show the balance in the conversation between the human and the bot, and the complexity of words used by it. We compare conversations from the chatbot with those from ChatGPT and Gemini where we found the last two to be too complex for a beginner learner.