跳到正文
TakoAPI
返回技能列表

install-llm-council

由 jeadland 提供

Web & Frontend Development
—暂无评分
你的评分

简介

LLM Council — multi-model consensus app with one-command setup.

描述

该技能的功能

LLM Council runs one question through multiple AI models, has them critique each other's answers, and produces a synthesized consensus response. A single command clones the repo, installs dependencies, resolves credentials, and starts a React/Vite frontend with a FastAPI backend. No manual API key prompts — credentials are read from environment, workspace .env, or the OpenClaw local gateway automatically.

Unlike manually querying multiple model APIs and reconciling results yourself, the chairman synthesis step produces a single reconciled answer from all participating models in one UI.

何时使用

  • Comparing how different models answer a disputed technical question
  • Getting a synthesized consensus before making an architectural decision
  • Exploring where models diverge on ambiguous or opinion-heavy prompts
  • Running a multi-model critique session without configuring each provider manually
  • Replacing a single-model chat with a panel review for high-stakes writing

示例工作流

输入

User runs: /install-llm-council

智能体
  1. 1

    Run install.sh from the skill directory

  2. 2

    Script resolves OPENROUTER_API_KEY from env, workspace .env, or OpenClaw gateway token

  3. 3

    Clone or update jeadland/llm-council repo to ~/workspace/llm-council

  4. 4

    Run uv sync for Python backend deps and npm ci for frontend, then write .env

  5. 5

    Start FastAPI backend on port 8001 and Vite dev server on port 5173 with health checks

输出

LLM Council running at http://10.0.1.42:5173 — user opens browser and submits a question to multiple models

前置要求

  • OpenRouter API key (only if not using OpenClaw gateway)

安装

clawhub install jeadland/install-llm-council

运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。

链接