User-controlled Multi-Agent Conversation (Human-in-the-loop)

Apologies for reposting this feature request here; due to my inexperience, I initially posted it in the Ideas section. Sorry for the inconvenience.

I am not sure if other software already implements a solution like this, but as far as the most well-known ones are concerned, I haven't seen it yet. From a programming standpoint, implementation should not be overly complex.

The core idea is based on the fact that every LLM has unique strengths and weaknesses. By making them interact, these limitations can be mitigated, significantly enhancing the quality of the conversation (hallucination mitigation) and the depth of the output.
The concept involves a chat interface where the user acts as a moderator, 'inviting' two or more LLMs to participate in one conversation. The user would establish the dialogue's rules from the beginning. For example (only):

Manual Moderation: The user sets the conversation goals, grants the floor to Model A, and decides when (or if) it is Model B's turn.

Fixed Rotation: A set sequence such as 1. User, 2. Model A, 3. Model B, and repeat.

Autonomous Interaction: The user sets the objective and allows the models to interact with each other autonomously until the user decides to interrupt.

Of course, this feature would be contingent on having enough RAM and VRAM to load multiple models into memory simultaneously.

I believe this would be a truly innovative and powerful feature!

No files, tests, or entry points are identified in the request. Start by reviewing LocalAI’s existing chat and model-orchestration entry points, then clarify requirements for manual moderation, fixed rotation, and autonomous interaction. Done requires agreed behavior and acceptance criteria for multi-model conversations, including interruption and resource constraints.