Colloquium
Colloquium is a multi-agent deliberation platform where AI personas with distinct expertise debate hypotheses through emergent, energy-driven dynamics.
Core trick
Persona agents self-organize into phases using energy-driven dynamics rather than scripted turn order, with red-team agents challenging assumptions.
Description
Colloquium is a multi-agent deliberation platform where AI personas with distinct expertise debate hypotheses through emergent, energy-driven dynamics. Agents self-organize into phases, challenge assumptions via red-teaming, and converge on consensus - mimicking how expert panels reason through complex scientific questions.
README excerpt
# Colloquium **Emergent multi-agent deliberation — where complex scientific discourse arises from simple rules, not engineered choreography.** [](https://github.com/sunitj/Colloquip/actions/workflows/ci.yml) [](https://www.python.org/downloads/) [](LICENSE) [](#testing) > *"Complex behavior emerges from simple rules."* — Stephen Wolfram Colloquium is a full-stack multi-agent deliberation platform where AI agents with distinct scientific personas debate hypotheses through **self-organizing phases**. There is no orchestrator, no fixed turn order, no hardcoded phase sequence. Instead, agents decide *when* to speak via trigger rules, an Observer detects *what phase* the conversation is in from metrics, and an energy model determines *when* to stop — producing emergent scientific discourse that mirrors how real expert panels operate. --- ## What Makes This Different | Traditional Multi-Agent | Colloquium (Emergent) | |---|---| | Fixed turn order (A → B → C → repeat) | Agents self-select when to speak via 9 trigger rules | | Predefined phase schedule | Observer detects phases from conversation dynamics | | Hard turn limit or manual stop | Energy-based termination — conversation dies nat