Docs › LLM Proxy

LLM Proxy Gateway

Payment wires: every paid endpoint accepts x402 and MPP (Machine Payments Protocol) on the same 402 - see Paying with x402 and Paying with MPP. Agent402 is the applied layer of Agentic Finance: agents that pay and get paid on their own.

Acceptable use. Hosted-instance traffic is governed by the Terms of Service - including a generative-content acceptable-use policy - and by the upstream model providers' usage policies. Wallets used for prohibited content are blocked before settlement. Outputs are generated by third-party models from your inputs; you are responsible for how you use them.

Looking for the OpenAI-compatible wire paths? The LLM Gateway (OpenAI /v1) serves standard chat/completions, embeddings, rerank, images/generations, videos/generations and audio/speech at https://agent402.tools/v1 (plus the Anthropic Messages and OpenAI Responses wires) - flat chat tiers from $0.003 per call, a metered tier quoted per request, embeddings and rerank from $0.002, model-optional auto-routing, streaming, and response caching. Any OpenAI SDK adopts it by changing base_url. The /api/llm* tools below remain available as the simpler custom-JSON proxy.

Three tiers of OpenAI inference, paywalled via x402. An agent sends a standard chat/completions request and pays per call over x402 (USDC on Base and the other rails its 402 lists, or USDG on Robinhood Chain) or over MPP. No model API key is needed on the caller's side needed on the caller's side, no signup, no account. The operator's own OPENAI_API_KEY handles upstream auth; per-tier input and output caps bound what one call can ask for.

Tiers

Endpoint	Price	Models	Input cap	Output cap
`POST /api/llm`	$0.01	`gpt-4o-mini`	16k chars	4,096 tokens
`POST /api/llm-pro`	$0.10	`gpt-4o`, `gpt-4.1`	16k chars	2,048 tokens
`POST /api/llm-premium`	$0.50	`gpt-5.6-terra` (default), `o3` (reasoning)	32k chars	2,048 tokens

GPT-4o, GPT-4o mini and GPT-4.1 are scheduled to be retired upstream on 2027-01-14; these routes will move to their successors before then.

All three tiers are wallet-only: there is no proof-of-work free tier because every call burns real upstream inference credit. See Security Model for the wallet-only rationale.

Request format

Every tier accepts the same OpenAI-shaped JSON body:

{
  "model": "gpt-4o-mini",
  "messages": [
    { "role": "system", "content": "You are a concise assistant." },
    { "role": "user", "content": "Summarize x402 in two sentences." }
  ],
  "max_tokens": 256,
  "temperature": 0.7,
  "top_p": 1,
  "stop": "\n"
}

Field	Required	Notes
`model`	no	Defaults to `gpt-4o-mini` on `/api/llm` and `/api/llm-pro`, and to `gpt-5.6-terra` on `/api/llm-premium`. Must match the tier's allowlist by prefix (`gpt-4o-mini` for `/api/llm`; `gpt-4o` / `gpt-4.1` for `/api/llm-pro`; `o3` / `gpt-5.6-terra` for `/api/llm-premium`, where the retired `o3-mini` and `o4-mini` ids are served by `gpt-5.6-terra`)
`messages`	yes	Array of `{role, content}` objects; max 50 messages
`max_tokens`	no	Default 1,024; silently clamped to the tier's output cap
`temperature`	no	0--2; controls randomness
`top_p`	no	0--1; nucleus sampling
`stop`	no	Stop sequence(s)

Response format

Every tier returns the same OpenAI-shaped envelope:

{
  "model": "gpt-4o-mini-2025-07-18",
  "provider": "openai",
  "usage": {
    "prompt_tokens": 24,
    "completion_tokens": 31,
    "total_tokens": 55
  },
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "x402 is an open HTTP payment standard that uses the 402 status code to let servers charge per request, settled on-chain in USDC. Buyers need only a wallet -- no API key, no account."
      },
      "finish_reason": "stop"
    }
  ]
}

Curl examples

Basic tier ($0.01: GPT-4o-mini)

curl -X POST https://agent402.tools/api/llm \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [{"role": "user", "content": "Say hello in one sentence."}],
    "max_tokens": 64
  }'

Pro tier ($0.10: GPT-4o / GPT-4.1)

curl -X POST https://agent402.tools/api/llm-pro \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "gpt-4o",
    "messages": [{"role": "user", "content": "Explain proof-of-work in three sentences."}],
    "max_tokens": 256
  }'

Premium tier ($0.50: gpt-5.6-terra / o3 reasoning)

curl -X POST https://agent402.tools/api/llm-premium \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "gpt-5.6-terra",
    "messages": [{"role": "user", "content": "What are the trade-offs of on-chain vs off-chain settlement?"}],
    "max_tokens": 512
  }'

All three return 402 without a valid x402 payment header: the same flow as every other paid tool. See Paying with x402 for how to sign and attach payment.

Per-tier caps

Each tier bounds what a single call can ask for:

Input cap (chars) limits how much prompt text the caller can send, bounding prompt-token cost.
Output cap (tokens) hard-limits max_completion_tokens sent to OpenAI, bounding completion-token cost.
Model allowlist per tier keeps each endpoint to the models its tier serves (e.g. sending gpt-4o to /api/llm returns 400).
Message count is capped at 50 per request.

These caps are enforced server-side before the upstream call is made.

Environment gating

The kit reads OPENAI_API_KEY at call time, not at boot. If the key is missing or empty, calls return 503 Service Unavailable and the rest of the server continues running normally. Self-hosters who don't want to offer LLM proxy simply omit the env var: no code change needed.