Conclave is a multi‑agent decision engine that runs structured debates between specialized agents (Planner, Skeptic, Security, Cost, Synthesizer). Instead of returning one answer, it produces competing proposals, a probability distribution over them, and a final response synthesized from the debate. When a prompt involves code, Conclave can attach verification evidence from a local MCP tool chain.
- Generates distinct agent proposals and a ranked probability distribution (never 0%/100%).
- Shows explicit disagreement, risks, and assumptions.
- Produces a final response (not just a recommendation), derived from the debate.
- Optionally runs verification for code prompts and surfaces evidence logs.
- Vite + React UI for debate inputs, weights, and outputs.
- Lightweight Node server (
/server) orchestrates debate runs and persists sessions. - LeanMCP server (
/mcp/conclave-tools) exposes verification tools (ping,verify_plan). - LLM providers: Local (Ollama) or API (Groq), selectable in the UI.
- Copy
.env.exampleto.env. - Set LLM provider vars:
- Ollama:
LLM_PROVIDER=ollama,OLLAMA_URL,OLLAMA_MODEL - Groq:
LLM_PROVIDER=groq,GROQ_URL,GROQ_API_KEY,GROQ_MODEL
- Ollama:
- Start MCP tools:
npm run dev:mcp - Start API:
npm run dev:server - Start UI:
npm run dev - Or run all:
npm run dev:full
- Verification runs only for code prompts.
- If MCP is unreachable, the server falls back to mock verification and records a warning.
- The UI notes when verification is skipped for non‑code prompts.
- The “Final Diff” is a demo artifact (not a real patch).
- Evidence entries are ordered by the final probabilities.