API documentation

Base URL and formats

Every request goes to https://api.klinch.cloud. One key works with every model and both API formats:

Any model answers in either format: call gpt-6.1-sol with the Anthropic SDK or claude-sonnet-5-5 with the OpenAI SDK. The Anthropic SDK takes the base URL without /v1, the OpenAI SDK with it.

Keys and authentication

A key looks like kl_live_… and arrives in the FunPay chat right after you buy a token pack. Add it to the klinch.app cabinet to see the balance, spend and request history.

Send the key in either header; both work on every endpoint:

• Authorization: Bearer kl_live_… (what the OpenAI SDK sends)
• x-api-key: kl_live_… (what the Anthropic SDK sends)

First request

curl https://api.klinch.cloud/v1/messages \
  -H "x-api-key: kl_live_…" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-5-5",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

The reply, in the Anthropic format:

{
  "id": "msg_01…",
  "type": "message",
  "role": "assistant",
  "model": "claude-sonnet-5-5",
  "content": [{"type": "text", "text": "Hello! How can I help?"}],
  "stop_reason": "end_turn",
  "usage": {"input_tokens": 9, "output_tokens": 12}
}

usage holds the real tokens of the request. The balance is charged that number times the model’s rate (see models and pricing).

Anthropic Messages API

The Anthropic API fields work: system, messages, max_tokens, temperature, tools, tool_choice, thinking, stream, images and documents in content, and caching with cache_control.

Streaming in Python:

import anthropic

client = anthropic.Anthropic(base_url="https://api.klinch.cloud", api_key="kl_live_…")
with client.messages.stream(
    model="claude-sonnet-5-5", max_tokens=1024,
    messages=[{"role": "user", "content": "Write a haiku"}],
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

KLINCH adds prompt caching by itself when it pays off (a long system prompt, tools, a conversation history). Cache reads cost a tenth of regular input. To opt out, send "cache_control": null in the body.

OpenAI Chat Completions

For the OpenAI SDK, LangChain, LlamaIndex, Cursor and any client with a custom base_url. messages, stream, tools, response_format (JSON), temperature and max_tokens are supported.

from openai import OpenAI

client = OpenAI(base_url="https://api.klinch.cloud/v1", api_key="kl_live_…")
r = client.chat.completions.create(
    model="gpt-6.1-sol",
    messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)

Streaming in Node.js:

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.klinch.cloud/v1", apiKey: "kl_live_…" });
const stream = await client.chat.completions.create({
  model: "deepseek-v4-pro",
  messages: [{ role: "user", content: "Explain recursion" }],
  stream: true,
});
for await (const chunk of stream) process.stdout.write(chunk.choices[0]?.delta?.content || "");

OpenAI Responses API

The format Codex CLI and newer OpenAI SDKs use. input is a string or a list of messages; stream and tools are supported.

curl https://api.klinch.cloud/v1/responses \
  -H "Authorization: Bearer kl_live_…" \
  -H "content-type: application/json" \
  -d '{"model": "gpt-6.1-sol", "input": "Hello"}'

Tools (function calling)

Tools work in both formats and with every model. The model returns a call, you run the function and send the result in the next message, just like the official APIs.

r = client.chat.completions.create(
    model="claude-sonnet-5-5",
    messages=[{"role": "user", "content": "Weather in Almaty?"}],
    tools=[{
        "type": "function",
        "function": {
            "name": "get_weather",
            "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]},
        },
    }],
)
print(r.choices[0].message.tool_calls)

List models

The models your key can use right now, with their balance rates:

curl https://api.klinch.cloud/v1/models -H "Authorization: Bearer kl_live_…"

{
  "object": "list",
  "data": [
    {
      "id": "claude-sonnet-5-5",
      "object": "model",
      "display_name": "Claude Sonnet 5.5",
      "owned_by": "anthropic",
      "multipliers": {"input": 1, "output": 5, "cache_read": 0.1, "cache_write": 1.25}
    }
  ],
  "has_more": false
}

Descriptions and prices of every model are on Models and pricing.

Count tokens

How many input tokens a request will take, before you send it. Free.

curl https://api.klinch.cloud/v1/messages/count_tokens \
  -H "x-api-key: kl_live_…" \
  -H "content-type: application/json" \
  -d '{"model": "claude-sonnet-5-5", "messages": [{"role": "user", "content": "Hello"}]}'

Embeddings

POST /v1/embeddings in the OpenAI format. The default model is bge-m3 (multilingual, 1024 dimensions). OpenAI’s model names (text-embedding-3-small and the rest) are accepted too. Model list: GET /v1/embeddings/models.

r = client.embeddings.create(model="bge-m3", input=["first text", "second text"])
print(len(r.data[0].embedding))  # 1024

Text to speech

POST /v1/audio/speech in the OpenAI format. With the model tts-1 KLINCH picks a voice engine for the language of the text. Formats: mp3, opus, aac, flac, wav, pcm. Up to 4096 characters per request. Models and voices: GET /v1/audio/models.

r = client.audio.speech.create(model="tts-1", voice="alloy", input="Привет! Это KLINCH.")
r.write_to_file("hello.mp3")

Balance and usage

The balance is counted in tokens. 1 balance token equals one Claude Sonnet 5.5 input token; other models charge more or less by their rate.

curl https://api.klinch.cloud/v1/balance -H "Authorization: Bearer kl_live_…"

{
  "id": 123,
  "balance": 1250000,
  "unit": "tokens",
  "key_prefix": "kl_live_ab12"
}

• GET /v1/usage?days=30: requests, tokens and spend by day and model.
• GET /v1/history: recent requests with tokens and charges.
• GET /v1/limits: the key’s usage limits and when they reset.

Errors

Errors come with an HTTP status and a JSON body. On the OpenAI endpoints the body follows OpenAI’s error shape.

{"error": "insufficient_balance", "required": 5200, "balance": 310, "unit": "tokens"}

Limits

Up to 600 requests a minute per IP address, plenty for Claude Code, agents and bots. When a reply would cost more than the balance, KLINCH does not fail the request: it lowers max_tokens so the balance never goes negative.

The status page shows how the models work in real time.

Claude Code, Cursor and other tools

Claude Code needs two environment variables:

export ANTHROPIC_BASE_URL="https://api.klinch.cloud"
export ANTHROPIC_AUTH_TOKEN="kl_live_…"
claude

Step-by-step guides for other tools, Windows and macOS are in the cabinet under Connect, and on the Claude Code and Cursor pages.