moa
Mixture of Agents: Make 3 frontier models argue, then synthesize their best insights into one superior answer. ~$0.03/query.
- License 许可
- MIT
- 最近更新
- 2026-06-07
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- cost
- ~$0.03 per query (paid tier)
- name
- moa
- author
- John Scianna (@Scianna)
- version
- 1.2.0
- requires
- [ "OPENROUTER_API_KEY" ]
- description
- Mixture of Agents: Make 3 frontier models argue, then synthesize their best insights into one superior answer. ~$0.03/query.
03 SKILL.md 正文
Mixture of Agents (MoA)
TL;DR: Make 3 AI models argue with each other. Get an answer better than any single model. Cost: ~$0.03.
Two Usage Modes
A. Standalone CLI (Node.js)
export OPENROUTER_API_KEY="your-key"
node scripts/moa.js "Your complex question"
B. OpenClaw Skill (Agent-orchestrated)
# Install
clawhub install moa
# Or copy to ~/clawd/skills/moa/
The agent can then invoke MoA for complex analysis tasks.
Origin Story
The concept of "Mixture of Agents" comes from research showing LLMs can improve each other's outputs through collaboration. I built this for VC deal analysis—when evaluating startups, you want multiple perspectives, not one model's opinion.
The journey:
- Started with 5 free OpenRouter models (Llama, Gemini, Mistral, Qwen, Nemotron)
- Rate limits killed me at 2am during peak hours
- Switched to 3 paid frontier specialists
- Result: ~$0.03/query, answers better than any single model
When to Use
- Complex analysis — due diligence, market research, technical evaluation
- Brainstorming — get diverse ideas, synthesize the best
- Fact-checking — cross-reference across models with different training data
- High-stakes decisions — when one model's blind spots could hurt you
- Contrarian thinking — different models have different biases
When NOT to use:
- Quick Q&A (too slow, 30-90s latency)
- Real-time chat (not designed for streaming)
- Simple lookups (overkill)
Model Configuration
Paid Tier (Default) — Recommended
| Role | Model | ~Latency | Strength |
|---|---|---|---|
| Proposer 1 | moonshotai/kimi-k2.5 |
23s | Long context, strong reasoning |
| Proposer 2 | z-ai/glm-5 |
36s | Technical depth, different training corpus |
| Proposer 3 | minimax/minimax-m2.5 |
64s | Nuance catching, thorough analysis |
| Aggregator | moonshotai/kimi-k2.5 |
15s | Fast synthesis |
Why these models?
- Frontier-class but less congested than GPT-4/Claude
- Different training data = genuinely different perspectives
- Chinese models excel at certain reasoning tasks
- Combined cost still cheaper than single Opus call
Cost breakdown:
3 proposers × ~$0.008 = $0.024
1 aggregator × ~$0.005 = $0.005
─────────────────────────────
Total: ~$0.029/query
Free Tier (Fallback)
5 models: Llama 3.3 70B, Gemini 2.0 Flash, Mistral Small, Nemotron 70B, Qwen 2.5 72B
⚠️ Warning: Free tier hits rate limits during peak hours. Use --free flag only for testing.
How It Works
┌─────────────┐
│ PROMPT │
└──────┬──────┘
│
┌──────────┼──────────┐
▼ ▼ ▼
┌────────┐ ┌────────┐ ┌────────┐
│Kimi 2.5│ │ GLM 5 │ │MiniMax │ ← Parallel (they "argue")
│(reason)│ │(depth) │ │(nuance)│
└───┬────┘ └───┬────┘ └───┬────┘
│ │ │
└──────────┼──────────┘
▼
┌──────────────┐
│ AGGREGATOR │
│ (Kimi 2.5) │
│ │
│ • Best of 3 │
│ • Resolve │
│ conflicts │
│ • Synthesize │
└──────┬───────┘
▼
┌──────────────┐
│ FINAL ANSWER │
│ (Synthesized)│
└──────────────┘
API Reference
Function Signature
interface MoAOptions {
prompt: string; // Required: The question to analyze
tier?: 'paid' | 'free'; // Default: 'paid'
}
interface MoAResult {
synthesis: string; // The final aggregated answer
}
// Throws on complete failure (all models down, invalid key)
// Returns partial synthesis if 1-2 models fail
async function handle(options: MoAOptions): Promise<string>
CLI Usage
# Paid tier (default)
node scripts/moa.js "Your complex question"
# Free tier
node scripts/moa.js "Your question" --free
Programmatic Usage
const { handle } = require('./scripts/moa.js');
const synthesis = await handle({
prompt: "Analyze the competitive moats in AI code generation",
tier: 'paid'
});
console.log(synthesis);
Failure Modes
| Scenario | Behavior |
|---|---|
| 1 proposer fails | Synthesis from remaining 2 models |
| 2 proposers fail | Synthesis from 1 model (degraded) |
| All proposers fail | Returns error message |
| Invalid API key | Immediate error with setup instructions |
| Rate limit (free tier) | Returns rate limit error |
The system is designed to degrade gracefully. A 2/3 response is still valuable.
Example Use Cases
VC Due Diligence
node scripts/moa.js "Analyze the competitive landscape for AI code generation. \
Who has defensible moats? Who's likely to be commoditized? Be specific."
Technical Evaluation
node scripts/moa.js "Compare RLHF vs DPO vs RLAIF for LLM alignment. \
Which scales better? What are the failure modes of each?"
Market Research
node scripts/moa.js "What are the emerging use cases for embodied AI in 2026? \
Focus on robotics, drones, and autonomous systems. Include specific companies."
Performance Expectations
| Metric | Paid Tier | Free Tier |
|---|---|---|
| P50 Latency | ~45s | ~60s |
| P95 Latency | ~90s | ~120s+ |
| Success Rate | >99% | ~80% (rate limits) |
| Cost/Query | ~$0.03 | $0.00 |
Tips
- Be specific — Vague prompts get vague synthesis
- Ask for structure — "Give me pros/cons" or "List top 5" helps the aggregator
- Use for analysis, not chat — MoA shines for complex reasoning
- Batch your queries — 30-90s per query, so plan accordingly
Installation
Via ClawHub (Recommended)
clawhub install moa
Manual
- Copy
skills/moa/to your~/clawd/skills/directory - Set
OPENROUTER_API_KEYin your environment - The agent can now invoke MoA for complex queries
Environment Variables
| Variable | Required | Description |
|---|---|---|
OPENROUTER_API_KEY |
Yes | Your OpenRouter API key |
Get your key at: https://openrouter.ai/keys
Credits
- MoA concept: Together AI Research
- Implementation: @Scianna
- Built for: OpenClaw
04 同仓库更多 Skills
crowd-prompting
A marketplace where AI agents improve prompts, system instructions, tool descriptions, and other text-based content with domain expertise from real-world operations — and earn tokens for valuable contributions.
ceaser
Interact with the Ceaser privacy protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- shield, unshield, note management, and protocol queries. All ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). Ceaser is a privacy-preserving ETH wrapper using Noir/UltraHonk zero-knowledge proofs on Base L2 (chain ID 8453).
ceaser-send
Fully automated private ETH transfer via Ceaser Protocol on Base L2 using the ceaser-mcp MCP tools. This skill uses the ceaser-mcp npm package for ALL operations -- all ceaser tool calls use CLI subcommands (npx -y ceaser-mcp <subcommand>). WARNING -- creates observable on-chain links between user wallet and hot wallet (see Privacy Warning). Generates an ephemeral hot wallet with BIP-39 mnemonic recovery, user funds it, agent signs and broadcasts the Shield TX automatically. Exactly one manual step required (funding the hot wallet with ETH). Uses Noir/UltraHonk zero-knowledge proofs.
dilemm-ai
Compete in DilemmAI, the prisoner's dilemma AI arena at dilemm.ai. Use when an OpenClaw agent wants to create an account, design and submit strategy prompts for their bot, enter matchmaking, analyze historical matches and leaderboards, and iterate on bot strategy. Requires headless browser automation (no API available yet).
moldium
Post and manage content on the Moldium blog platform. Triggered by "post to Moldium", "write a blog post", "publish an article", etc.
arxiv-paper-reviews
Interact with arXiv Crawler API to fetch papers, read reviews, submit comments, search papers, and import papers. Use when working with arXiv papers, fetching paper lists by date/category/interest, viewing paper details with comments, submitting paper reviews, searching papers by title, or importing papers from arXiv URLs via API at http://weakaccept.top:8000/.