Build on useGlimper.

1. Buy credits and copy your key. 2. Set GLIMPER_KEY. 3. Point your SDK at the base URL.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://useglimper.fun/v1",
    api_key=os.environ["GLIMPER_KEY"],
)

res = client.chat.completions.create(
    model="glimper/dolphin-mistral-24b",
    messages=[{"role": "user", "content": "Ask anything."}],
)
print(res.choices[0].message.content)

Send your key in the Authorization header. Keys are shown once. We only store a hash. Rotate them any time from Account.

Authorization: Bearer sk-glimper-XXXXXXXXXXXXXXXX

POST /v1/chat/completions takes the same parameters as OpenAI: model, messages, temperature, max_tokens, tools, response_format.

{
  "id": "chatcmpl-9x2…",
  "model": "glimper/hermes-4-405b",
  "upstream_model": "nousresearch/hermes-4-405b",
  "choices": [{ "message": { "role": "assistant", "content": "…" } }],
  "usage": { "prompt_tokens": 18, "completion_tokens": 212, "total_tokens": 230, "cost_usd": 0.000654 }
}

Every response reports the exact upstream_model and the cost_usd.

Set "stream": true to receive server-sent events. Gateway overhead is under 40 ms p50.

curl https://useglimper.fun/v1/chat/completions -N \
  -H "Authorization: Bearer $GLIMPER_KEY" -H "Content-Type: application/json" \
  -d '{"model":"glimper/dolphin-mistral-24b","stream":true,
       "messages":[{"role":"user","content":"Tell me a dark story."}]}'

POST /v1/images/generations with an image model such as glimper/lustify-sdxl. Sizes: square, portrait, landscape; up to 4 images per request, billed per image returned.

curl https://useglimper.fun/v1/images/generations \
  -H "Authorization: Bearer $GLIMPER_KEY" -H "Content-Type: application/json" \
  -d '{"model":"glimper/chroma","prompt":"noir city at 3am, rain, neon","size":"portrait","n":1}'
# → { "data": [{ "b64_json": "…" }], "cost_usd": 0.012 }

Prompts involving minors are refused for every model.

GET /v1/models returns every model with its price per 1M tokens. Filter with ?uncensored=true.

Use useGlimper as the model provider in any agent that supports OpenAI-compatible endpoints: Claude Code proxies, Cursor, Cline, OpenWebUI, SillyTavern, LiteLLM.

# LiteLLM / OpenWebUI / Cline
OPENAI_API_BASE=https://useglimper.fun/v1
OPENAI_API_KEY=$GLIMPER_KEY
MODEL=glimper/hermes-4-405b

No account and no API key: call /v1/chat/completions without an Authorization header and pay each request in USDG on Robinhood Chain with x402. The server answers 402 with a PAYMENT-REQUIRED header quoting the most the request can cost; your client signs a USDG transferWithAuthorization for it and retries with PAYMENT-SIGNATURE. useGlimper pays the gas.

You are billed the real token cost. Whatever is left of the payment stays as credit for your wallet and pays for its next requests, so most calls after the first settle nothing on chain.

# network  eip155:4663 (Robinhood Chain)
# asset    USDG 0x5fc5360D0400a0Fd4f2af552ADD042D716F1d168
# scheme   exact (EIP-3009), EIP-712 domain "Global Dollar", version "1"

import { wrapFetchWithPaymentFromConfig } from "@x402/fetch";
import { ExactEvmScheme } from "@x402/evm";
import { privateKeyToAccount } from "viem/accounts";

const USDG = "0x5fc5360D0400a0Fd4f2af552ADD042D716F1d168";
const pay = wrapFetchWithPaymentFromConfig(fetch, {
  schemes: [{ network: "eip155:4663", client: new ExactEvmScheme(privateKeyToAccount(KEY)) }],
  spendControls: { allowedAssets: [{ network: "eip155:4663", asset: USDG, maxAmountPerPayment: "1000000" }] },
});
const res = await pay("https://useglimper.fun/v1/chat/completions", {
  method: "POST",
  headers: { "content-type": "application/json" },
  body: JSON.stringify({ model: "glimper/hermes-4-405b", max_tokens: 800, messages: [{ role: "user", content: "hi" }] }),
});

max_tokens sets the quote (default 2,048). The settlement transaction comes back in the PAYMENT-RESPONSE header.

Every completion returns a receipt signed by useGlimper's server wallet: model, token counts, price, time and the keccak256 of the generated text. The text itself is never stored.

Non-streaming responses carry it in a receipt field and the x-glimper-receipt header. Streams end with an SSE comment, : glimper-receipt <base64>, which OpenAI clients ignore. Each UTC day's receipts are folded into one Merkle root and written to Robinhood Chain. Check any receipt at /verify, or fetch its proof from GET /api/receipts/:id?h=<outputHash>.

max_tokens is capped to what your balance can pay for. Requests are rejected with 402 when the balance runs out.