council-of-llms

What problem does it solve?

Single-model “councils” that roleplay multiple experts often produce shallow analysis and inconsistent outcomes, because they lack genuinely independent perspectives and adequate context control.

Core Features & Use Cases

• Parallel multi-model deliberation: Spawns three subagents on different models aligned to distinct analytical lenses (Strategic, Analytical, Creative) so their outputs are independently generated.
• Unified synthesis into a verdict: Merges results into consensus points, disagreements (with why), blind spots, a final weighted verdict, and concrete action items.
• Practical guardrails against common failures: Encourages inline context, lightContext execution, prompt size limits, and timeout handling to reduce empty outputs and context overflow.
• Use Case: Evaluate a high-stakes plan (e.g., launching an automation workflow or refactoring an architecture) and produce a decision memo with risks and next steps drawn from three distinct perspectives.

Quick Start

Configure your three models in ~/.openclaw/council-config.json, then run a council session by providing the full topic context inline to the three spawned subagents.