However, traditional single-agent chatbots are susceptible to technological hallucination, non-consistent pedagogical depth, and lack of a verification system, as they usually offer ready-made answers without promoting conceptual thinking. This paper suggests a Collaborative Multi-Agent Chatbot Framework that will facilitate intelligent tutoring and feedback of high quality for Java programming. In particular, we propose to use a modular design with specialized agents, such as a generative tutor agent and a verification agent who is tasked with verifying the generated answers. The verifier uses real-time execution of JUnit 5 tests to validate hints generated by other agents. As far as we know, this is one of the first attempts to use test-based verification in the process of LLM tutoring. The proposed system was tested using 15 selected Java programming problems that covered basic algorithms and object-oriented programming principles. The experimental findings indicate that the new method is considerably superior to the single-agent models, with 90% success rate. Although the multi-agent architecture may result in longer response times, it requires fewer interaction iterations, thus providing a more productive and educational experience.