for agents

Your agent can pay for what it runs.

Media generation, chat completions, and a searchable prompt library — one paid endpoint and one MCP server, on two rails: per-call USDC settlement with no account, or an API key that bills prepaid credits.

2

payment rails — x402 wallet or lp_ API key

11

tools on one MCP server — free and paid

$0.002

minimum paid call, metered per use

What an agent can call

Media generation

Text-to-image, image edit, text-to-video, and image-to-video across the hosted catalog — including a members-only private line. The quote arrives per result, before anything settles.

Chat completions

A standard chat-completions request shape against nine frontier and open-weight models. Billed at your max_tokens cap — set it tight to bill tight.

Prompt library

Search, enhance, trending, and packs built from real published results — every card carries the exact instruction and model that made it. Free tools over MCP.

Two ways to pay

x402 — no account

POST the endpoint; it answers 402 with a PAYMENT-REQUIRED quote. Your wallet settles it in USDC on Base mainnet; retry with the PAYMENT-SIGNATURE header. Pass payer = your wallet for settled-volume tiers — 5%, 10%, or 15% off at $100, $500, or $2,000 lifetime.

API key — prepaid credits

Sign up, buy credits by card or crypto, and create an lp_… key in settings. Send it as Authorization: Bearer lp_… — calls bill the prepaid balance, no 402 flow.

Invalid input, policy rejection, and provider failure never settle or bill — an agent only pays for a delivered result. The same guarantee is written into the manifest your agent reads.

Text models — live catalog

GPT-6 Lunaopenai/gpt-6-luna$0.15 in · $0.77 out /Mtok · 1.05M ctx
GLM 5.3 Flashz-ai/glm-5.3-flash$0.23 in · $0.77 out /Mtok · 1.31M ctx
Gemini 3.5 Flash Litegoogle/gemini-3.5-flash-lite$0.46 in · $3.85 out /Mtok · 1.05M ctx
Llama 3.3 70B — open weightsmeta-llama/llama-3.3-70b-instruct$1.11 in · $1.11 out /Mtok · 131k ctx
Claude Sonnet 5.5 — frontieranthropic/claude-sonnet-5.5$3.08 in · $15.38 out /Mtok · 1M ctx
DeepSeek 4.1 Flashdeepseek/deepseek-v4.1-flash$0.23 in · $0.92 out /Mtok · 1.05M ctx
Qwen3.7 Plusqwen/qwen3.7-plus$0.49 in · $1.97 out /Mtok · 1M ctx
Kimi K3moonshotai/kimi-k3$4.62 in · $23.08 out /Mtok · 1.05M ctx
Grok 4.20x-ai/grok-4.20$1.92 in · $3.85 out /Mtok · 2M ctx
Claude Opus 5.5anthropic/claude-opus-5.5$6.15 in · $30.77 out /Mtok · 1M ctx
Claude Fable 5.1anthropic/claude-fable-5.1$15.38 in · $76.92 out /Mtok · 1M ctx
Claude Haiku 4.5anthropic/claude-haiku-4.5$1.54 in · $7.69 out /Mtok · 200k ctx
DeepSeek R1 — reasoningdeepseek/deepseek-r1$1.08 in · $3.85 out /Mtok · 64k ctx
Gemini 3.8 Flashgoogle/gemini-3.8-flash$2.31 in · $11.54 out /Mtok · 1.05M ctx
Gemini 3 Flash Previewgoogle/gemini-3-flash-preview$0.77 in · $4.62 out /Mtok · 1.05M ctx
Gemma 4 31B — open weightsgoogle/gemma-4-31b-it$0.23 in · $0.62 out /Mtok · 262k ctx
Gemma 4 26B A4B — open weightsgoogle/gemma-4-26b-a4b-it$0.12 in · $0.49 out /Mtok · 262k ctx
MiniMax M3minimax/minimax-m3$0.46 in · $1.85 out /Mtok · 1.05M ctx
Mistral Largemistralai/mistral-large$3.08 in · $9.23 out /Mtok · 128k ctx
Mistral Nemomistralai/mistral-nemo$0.23 in · $0.23 out /Mtok · 131k ctx
Kimi K2.7 Codemoonshotai/kimi-k2.7-code$1.46 in · $6.15 out /Mtok · 262k ctx
GPT-6.1 Solopenai/gpt-6.1-sol$3.08 in · $15.38 out /Mtok · 1.05M ctx
GPT-6 Astraopenai/gpt-6-astra$15.38 in · $76.92 out /Mtok · 1.05M ctx
GPT-5.6 Terraopenai/gpt-5.6-terra$3.08 in · $18.46 out /Mtok · 1.05M ctx
GPT-5.5openai/gpt-5.5$7.69 in · $46.15 out /Mtok · 1.05M ctx
GPT-5.4 Miniopenai/gpt-5.4-mini$1.15 in · $6.92 out /Mtok · 400k ctx
GPT-5.4 Nanoopenai/gpt-5.4-nano$0.31 in · $1.92 out /Mtok · 400k ctx
GPT-5.3 Codexopenai/gpt-5.3-codex$2.69 in · $21.54 out /Mtok · 400k ctx
GPT-OSS 120B — open weightsopenai/gpt-oss-120b$0.23 in · $0.92 out /Mtok · 131k ctx
Qwen3.8 Flashqwen/qwen3.8-flash$0.23 in · $0.72 out /Mtok · 1M ctx
Qwen3.7 Maxqwen/qwen3.7-max$2.27 in · $6.81 out /Mtok · 1M ctx
Qwen3 Coder Nextqwen/qwen3-coder-next$0.46 in · $2.31 out /Mtok · 262k ctx
GLM 5.3z-ai/glm-5.3$2.15 in · $6.77 out /Mtok · 1.31M ctx

The 402 quote = estimated input tokens × in-rate + your max_tokens cap (≤ 8,192) × out-rate. The same table, machine-readable, at /v1/agent/models.

One MCP server

{
  "mcpServers": {
    "livepair": {
      "url": "https://livepairai.com/mcp"
    }
  }
}

Free tools work the moment it attaches — prompt search, trending, enhance, model listing, price quotes. Paid tools quote in USDC and settle the same way. Per-client config fields live in the MCP setup docs.

Drop-in integrations

Not building an agent? Ship a bot or an automation in minutes — every template is zero-dependency, open source, and bills your own API key:

Setup walkthroughs: Discord · n8n · CLI · Mobile apps — or deploy the Discord bot to Render in one click from the repo README.

Built under a judgment gate

This codebase doesn't ship on vibes. Before a change can deploy, a dedicated judgment model evaluates the diff against a fixed vocabulary of typed constraints — suppression checks, unverified delivery paths, silent-failure patterns, scope drift — all in one call, and only high-confidence verdicts can block.

Review memory persists between runs: files that fail get examined harder next time, and open obligations carry forward until resolved. A deterministic floor — dependency audits, endpoint vetting, contract checks — runs even when the judge is offline and can't be argued out of blocking.

It's the discipline agent pipelines need in general: the model proposes, typed rules dispose. The same gate audits this site's copy — which is why the numbers on this page match the running code.

The mechanics, in plain terms

Does my agent need an account or an API key?

Not on x402 — the caller's wallet settles each call in USDC on Base mainnet. The second rail is a standard API key: sign up, buy credits by card or crypto, create an lp_ key in settings, and send it as Authorization: Bearer lp_… — calls bill the prepaid balance.

What does a call cost?

Media is priced per result — the quote arrives with the 402 before anything settles. Chat is priced from token estimates: estimated input tokens at the model's in-rate plus your max_tokens cap (≤ 8,192) at the out-rate, with a $0.002 floor.

What happens when a generation fails?

Nothing settles. Invalid input, policy rejection, and provider failure all exit before settlement — you only pay for a delivered result.

Can an agent resell generation?

Yes — pass `payer` = your wallet and settled lifetime volume unlocks automatic discounts (5% at $100, 10% at $500, 15% at $2,000). Generated media is resellable at your own pricing.

Endpoints

MCP serverhttps://livepairai.com/mcpstreamable-http, no auth
Model catalogGET /v1/agent/modelsfree
ManifestGET /v1/agent/manifest.jsonfree
OpenAPI 3.1GET /v1/agent/openapi.jsonfree
GeneratePOST /v1/agent/generatepaid
ChatPOST /v1/agent/chatpaid
Job statusGET /v1/agent/jobs/{id}free
BalanceGET /v1/agent/balancelp_ key — prepaid balance + ledger
Machine guideGET /SKILL.mdfree

Point your agent at https://livepairai.com/mcp — free tools work on attach, paid tools settle in USDC per call. The full contract lives in the API reference and the MCP docs and the machine guide.