Teardowns of how real platforms orchestrate AI agents to facilitate group processes. Each captures agent roles, participant interaction models, stage pipelines, and data flow.
Inclusion criteria: Does an AI agent actively conduct, guide, or mediate a conversation or group process? Tools that only collect votes, structure arguments, or analyze data post-hoc are out of scope.
Source: workflows repo
Platforms
| Platform | Source | Type | Agents | Scale |
|---|---|---|---|---|
| Anthropic Interviewer | researched | Async 1-on-1 interviews | Interviewer + Analyst | 1,000+ |
| Bot Mediation | researched | AI dispute mediation | AI Bot Mediator | 2 (bilateral) |
| ComplexChaos | inferred | Async organizational alignment | Dialogue Agent + Pattern Synthesizer | 5-200 |
| CrowdSmart | researched | Generative Collective Intelligence | Collective Reasoning Agent + Pairwise Engine + Private LM | 10-10,000 |
| Habermas Machine | verified | Iterative consensus | Statement Generator + Critique Processor | 5-1,000 |
| Harmonica | verified | Async structured deliberation | Facilitator + Cross-Pollinator + Synthesizer | 3-500 |
| Juno | inferred | Unscripted AI interviews | Interviewer + Thematic Analyst | 5-1,000 |
| Listen | inferred | AI research pipeline | Recruiter + Interviewer + Analyst | 10-1,000 |
| Orchidea | researched | AI workshop facilitation | AI Ideator + Proposal Drafter | 3-500 |
| Outset | researched | Multimodal AI interviews | AI Moderator + Fraud Detector + Analyst | 10-5,000 |
| Parlay Ideas | researched | Classroom discussion facilitation | Parlay Genie + Participation Tracker | 5-40 |
| Remesh | inferred | Real-time collective dialogue | Clustering Engine + Consensus Predictor + Human Moderator | 20-5,000 |
| Talk to the City | verified | Elicitation + clustering | WhatsApp Bot + Topic Clusterer + Visualizer | 10-10,000 |
| Thinkscape | researched | Swarm intelligence | Conversational Surrogates + Orchestrator | 14-400 |
Source Types
- verified — documented from open-source code or detailed published methodology
- researched — documented from academic papers or detailed technical write-ups
- inferred — reconstructed from public marketing and website descriptions
Workflow Schema
Each YAML file describes:
- Agents — roles like facilitator, synthesizer, clustering engine, with type (
llm,human,hybrid,system) - Participants — input mode (text/voice/vote/mixed), interaction style, synchronicity, scale, anonymity
- Stages — ordered pipeline: input → processing → memory → output
- Patterns — which OFL facilitation patterns the platform implements (e.g., cross-pollination, Delphi)
See the schema definition for full details.
How This Fits into OFL
Teardowns are reference — a map of how the field actually builds AI facilitation today. They inform the method specs OFL is defining (portable protocols + their evals), alongside the evaluation frameworks and the rest of the knowledge base.