Docs

Quickstart

1. Request access. Onboarding is invite-only for now, and every account starts with 100 free credits.
2. Create an API key from the dashboard.
3. Send a request:

curl https://api.hakk.ai/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.6",
    "messages": [{"role": "user", "content": "Write a haiku about server racks."}]
  }'

Authentication

OpenAI-shaped clients send Authorization: Bearer YOUR_API_KEY. Claude Code can send the same key as x-api-key. Keys are created once on the API keys page and are scoped to your organization.

Anthropic clients may send anthropic-version and anthropic-beta. We accept those headers and do not require them.

Contract

Three public inference endpoints share one pipeline: auth, catalog, admission, credits, LiteLLM.

Accepted on Chat Completions:

• messages[].content as a string or as [{type, text}]
• max_tokens and max_completion_tokens
• developer role (mapped to system)
• tools, tool_choice, stream, temperature, stop, top_p
• Responses-shaped input and reasoning.effort on the completions path

Ignored extras stay ignored. Unknown fields do not 422.

We do not serve vision, embeddings, legacy /v1/completions, or Anthropic server tools.

OpenAI SDK

Python

from openai import OpenAI

client = OpenAI(base_url="https://api.hakk.ai/v1", api_key="YOUR_API_KEY")

response = client.chat.completions.create(
    model="qwen3.6",
    messages=[{"role": "user", "content": "hello"}],
)
print(response.choices[0].message.content)

# Codex CLI and newer agent code use Responses:
reply = client.responses.create(model="qwen3.6", input="hello")
print(reply.output[0].content[0].text)

JavaScript / TypeScript

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://api.hakk.ai/v1", apiKey: "YOUR_API_KEY" });

const response = await client.chat.completions.create({
  model: "qwen3.6",
  messages: [{ role: "user", content: "hello" }],
});
console.log(response.choices[0].message.content);

Streaming

Set "stream": true. Chat Completions uses OpenAI SSE frames ending in data: [DONE]. Messages uses Anthropic events (message_start, content_block_delta, ping, message_stop). Responses uses response.output_text.delta.

stream = client.chat.completions.create(
    model="qwen3.6",
    messages=[{"role": "user", "content": "hello"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

Editors & agents

Cursor

1. Open Cursor Settings (Cmd+Shift+J on macOS, Ctrl+Shift+J on Windows/Linux) and open Models.
2. Turn on OpenAI API Key and paste your hakk.ai key.
3. Turn on Override OpenAI Base URL and set it to https://api.hakk.ai/v1. Keep the /v1 suffix. Cursor appends /chat/completions to that URL.
4. Click Add Model and type each ID exactly. Cursor does not import our catalog when you refresh the model list. Use qwen3.6 for everyday work.
   • qwen3.6 (max thinking)
   • qwen3.6-fast (thinking off)
   • qwen3.6-high (faster thinking)
5. Select qwen3.6 in the chat model picker. If a request fails to connect, set Settings → Network → HTTP Compatibility Mode to HTTP/1.1.

Claude Code

Claude Code talks Anthropic Messages. Point it at the host root, not /v1. The client appends /v1/messages.

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.hakk.ai",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY",
    "ANTHROPIC_API_KEY": "",
    "ANTHROPIC_MODEL": "qwen3.6"
  }
}

Continue (VS Code / JetBrains)

models:
  - name: hakk.ai qwen3.6
    provider: openai
    model: qwen3.6
    apiBase: https://api.hakk.ai/v1
    apiKey: YOUR_API_KEY
    useResponsesApi: false

Cline

1. Open the Cline panel → Settings.
2. Set API Provider to "OpenAI Compatible".
3. Base URL https://api.hakk.ai/v1, your API key, model ID qwen3.6.

Aider

export OPENAI_API_BASE=https://api.hakk.ai/v1
export OPENAI_API_KEY=YOUR_API_KEY
aider --model openai/qwen3.6

Codex CLI

# ~/.codex/config.toml
[model_providers.hakk]
name = "hakk.ai"
base_url = "https://api.hakk.ai/v1"
env_key = "HAKK_API_KEY"
wire_api = "responses"

[models]
default = "qwen3.6"

Models

GET https://api.hakk.ai/v1/models lists the models your key can use. Type these IDs exactly in the editor. qwen-3.6 is accepted and mapped to qwen3.6.

Compatibility IDs still accepted: glm-5.2-fast, glm-5.2, glm-5.2-high, glm-5.3-high, glm-5.3-fast, glm-5.3. Same weights as the current model in that family.

The public catalog with per-1M-token prices is on the Models page, read from the same source billing charges against.

Credits & errors

Billing is prepaid credits (1 credit = $0.01), deducted per request from the model's published input/output token rates. When your organization's balance is exhausted, requests return 402. Editor routes also return {"error":{"message":"...","type":"...","code":"..."}} and keep detail as the same string. Top up on the billing page.

Under saturation you may briefly wait in the admission queue. If the wait exceeds the timeout you get 503 with a retry hint.