Skip to content

Conclave is a multi-agent decision engine that runs structured debates, scores competing plans, and delivers a final response with probability traces and evidence. Supports local LLMs or Groq API and optional code verification via MCP.

Notifications You must be signed in to change notification settings

the-axmc/conclave

Repository files navigation

Conclave

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.

What it does

  • 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.

Architecture

  • 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.

Run locally

  1. Copy .env.example to .env.
  2. Set LLM provider vars:
    • Ollama: LLM_PROVIDER=ollama, OLLAMA_URL, OLLAMA_MODEL
    • Groq: LLM_PROVIDER=groq, GROQ_URL, GROQ_API_KEY, GROQ_MODEL
  3. Start MCP tools: npm run dev:mcp
  4. Start API: npm run dev:server
  5. Start UI: npm run dev
  6. Or run all: npm run dev:full

Verification behavior

  • 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.

Notes

  • The “Final Diff” is a demo artifact (not a real patch).
  • Evidence entries are ordered by the final probabilities.

About

Conclave is a multi-agent decision engine that runs structured debates, scores competing plans, and delivers a final response with probability traces and evidence. Supports local LLMs or Groq API and optional code verification via MCP.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors