# Clervo x402 Gateway — Agent Instructions

You have access to Clervo, an OpenAI-compatible AI gateway on Base mainnet.
No API keys. No accounts. Free models need no payment. Paid models use x402 auto-payment.

## Base URL
https://api.clervo.dev

## Free Models (no payment, call directly)
- clervo/fast — 170ms, simple tasks (routes to Groq llama-3.1-8b)
- clervo/smart — 475ms, general reasoning (routes to Groq llama-3.3-70b)
- clervo/deep — ~1s, complex analysis (routes to Nvidia nemotron-ultra-550b)
- groq/llama-3.1-8b-instant — fastest, 8B params
- groq/llama-3.3-70b — best free quality, 70B params
- groq/qwen3.6-27b — Qwen 3.6, 27B params
- groq/gpt-oss-120b — GPT-OSS 120B
- groq/gpt-oss-20b — GPT-OSS 20B
- sambanova/llama-3.3-70b — alternate 70B route
- nvidia/nemotron-ultra-550b — largest free model, 550B MoE
- nvidia/nemotron-super-120b — 120B reasoning
- nvidia/nemotron-nano-30b — 30B efficient
- nvidia/deepseek-v4-flash — DeepSeek V4

## Paid Models ($0.005–$0.025 per call, x402 USDC on Base)
- clervo/code — Claude Sonnet 5, best for code ($0.015)
- quickai/gpt-5.4-mini — $0.005
- quickai/gpt-5.4 — $0.008
- quickai/gpt-5.5 — $0.012
- tongkhokr/claude-haiku-4.5 — $0.004
- tongkhokr/claude-sonnet-5 — $0.015
- tongkhokr/claude-opus-5 — $0.025

## Services

### Chat Completions
POST /v1/chat/completions
Content-Type: application/json

{"model":"clervo/fast","messages":[{"role":"user","content":"Hello"}],"max_completion_tokens":256}

### Web Search (free)
POST /v1/search
Content-Type: application/json

{"query":"your search query"}

Returns: {"results":[{"title":"...","url":"...","snippet":"..."}]}

### Web Scrape (free)
POST /v1/scrape
Content-Type: application/json

{"url":"https://example.com"}

Returns: {"content":"...markdown of page...","title":"..."}

### Audio Transcription (free)
POST /v1/audio/transcriptions
Content-Type: multipart/form-data

Send as multipart form with:
- file: audio file (mp3, mp4, wav, ogg, flac, webm, max 25MB)
- model: whisper-large-v3

Returns: {"text":"transcribed audio content"}

## Payment Flow (paid models only)
1. Send request → receive HTTP 402 with x-402-payment header (JSON)
2. Parse payment details: {network:"eip155:8453", token, amount, recipient, resource}
3. Sign USDC transfer on Base mainnet
4. Resend original request with x-402-receipt header containing signed payment
5. Receive normal 200 response with completion

## Tips
- Use clervo/fast for speed, clervo/smart for quality, clervo/deep for hard problems
- All responses follow OpenAI chat completions format
- max_completion_tokens controls output length (default varies by model)
- Free models have no rate limits during beta
- Search and scrape are always free, no payment needed