@certaworks/multi-voice-deliberation-framework
v0.1.0
Published
Adds structured Builder, Skeptic, Evolver, and Guardian debate to LLM pipelines before committing to an answer.
Maintainers
Readme
Multi-Voice Deliberation Framework
Type: SDK / Framework
Value: Adds structured Builder, Skeptic, Evolver, and Guardian debate to LLM pipelines before they commit to an answer.
Current Status
Complete as a local SDK / CLI / MCP slice. It can run a four-voice deliberation panel through an OpenAI-compatible adapter when local API credentials are provided.
Implemented Local Slice
- Builder, Skeptic, Evolver, and Guardian voice definitions with weighted confidence
deliberate()SDK orchestration with parallel or sequential voice calls- JSON response parsing with fallback handling for unstructured model output
- Synthesis text, aggregate concerns/suggestions, panel confidence, and human-review flag
- OpenAI-compatible adapter for OpenRouter, OpenAI, and Groq environment variables
deliberateCLI and stdio MCP server withdeliberateandlist_voicestools- Unit tests for voice orchestration, confidence, review thresholds, subsets, fallback parsing, and synthesis
Commands
npm install
npm test
npm run builddeliberate --prompt "Should this agent deploy the change?" --threshold 0.7 --jsonCurrent Limits
- This is a local product slice, not a hosted deliberation API or managed review service.
- No public npm publication, hosted accounts, API keys, billing, team workspace, live checkout, or web dashboard is included.
- Live deliberation requires one of
OPENROUTER_API_KEY,OPENAI_API_KEY, orGROQ_API_KEY. - The framework structures model calls and review signals; it does not guarantee factual correctness or safety.
Verification
Fresh suite verification on 2026-05-28:
npm testpassed, 10/10 tests.npm run buildpassed.
