Documentation

Wire your editor to the gateway.

Quickstart

Start with the hosted DarkCoder gateway at https://darkcoder.ai/v1. Create a revocable gateway key in the cockpit, then point any compatible client at that base URL.

curl https://darkcoder.ai/v1/models \
  -H "Authorization: Bearer dc_live_…"

Self-host

Prefer to operate the gateway yourself? Run DarkCoder on your own domain with the Node runtime and your own provider connections.

npm install
npm run dev
# Open http://localhost:3000/login

The development build seeds a local owner account: admin@darkcoder.ai / darkcoder.

The endpoint

Every request goes to one OpenAI-compatible surface: https://darkcoder.ai/v1. The model field accepts auto (the default combo), a combo slug like never-stop-coding, or a direct provider/model id like gh/gpt-5.5.

Cursor

Cursor talks to DarkCoder through its OpenAI-compatible override — no plugins, no login links.

DarkCoder → Cursor → Settings → Models

1. Enable "OpenAI API Key" and paste your DarkCoder key:
   dc_live_demo_cursor_home_0000000000003f9a
2. Set "Override OpenAI Base URL" to:
   https://darkcoder.ai/v1
3. Add a custom model named:
   auto

VS Code · Continue

# ~/.continue/config.yaml
models:
  - name: DarkCoder
    # DarkCoder speaks the OpenAI-compatible protocol — "openai" here is the protocol, not the vendor
    provider: openai
    model: auto
    apiBase: https://darkcoder.ai/v1
    apiKey: dc_live_demo_cursor_home_0000000000003f9a
    roles: [chat, edit, autocomplete]

Cline

DarkCoder → Cline → Settings → API Provider

Provider:        OpenAI Compatible
Base URL:        https://darkcoder.ai/v1
API Key:         dc_live_demo_cursor_home_0000000000003f9a (DarkCoder gateway key)
Model ID:        auto

curl

# DarkCoder via curl
curl -N https://darkcoder.ai/v1/chat/completions \
  -H "Authorization: Bearer dc_live_demo_cursor_home_0000000000003f9a" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto",
    "stream": true,
    "messages": [{ "role": "user", "content": "Explain this diff." }]
  }'

SDKs

DarkCoder via OpenAI SDK (Node)

// DarkCoder via OpenAI SDK (Node)
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://darkcoder.ai/v1",
  apiKey: "dc_live_demo_cursor_home_0000000000003f9a",
});

const res = await client.chat.completions.create({
  model: "auto",
  messages: [{ role: "user", content: "Hello" }],
});

DarkCoder via OpenAI SDK (Python)

# DarkCoder via OpenAI SDK (Python)
from openai import OpenAI

client = OpenAI(
    base_url="https://darkcoder.ai/v1",
    api_key="dc_live_demo_cursor_home_0000000000003f9a",
)

res = client.chat.completions.create(
    model="auto",
    messages=[{"role": "user", "content": "Hello"}],
)

Gateway keys

Sharing done right: invite a teammate with the base URL plus a dc_live_… gateway key you can revoke or rotate any time — never a Copilot login link. Raw provider tokens never leave this box; logs redact secrets automatically.

Combos

A combo is an ordered fallback chain with a strategy. The seeded never-stop-coding walks gh/gpt-5.5 → claude/claude-sonnet-4.5 → deepseek/deepseek-v3.2 → groq/llama-3.3-70b. Build your own in the cockpit under /app/combos.

Image generation

Connected image models are served through https://darkcoder.ai/v1/images/generations — OpenAI's images shape. Pass an image model id like oai/gpt-image-1, gemini/gemini-2.5-flash-image, grok/grok-2-image or a Higgsfield id like hf/soul-v2, or omit model to let the router pick the first connected image provider. Responses default to b64_json; response_format: "url" stores the image and returns a signed-in-only URL.

curl https://darkcoder.ai/v1/images/generations \
  -H "Authorization: Bearer dc_live_demo_cursor_home_0000000000003f9a" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "oai/gpt-image-1",
    "prompt": "A mint-green neon gateway logo on dark glass",
    "size": "1024x1024"
  }'
from openai import OpenAI

client = OpenAI(base_url="https://darkcoder.ai/v1", api_key="dc_live_demo_cursor_home_0000000000003f9a")
img = client.images.generate(
    model="auto",
    prompt="A mint-green neon gateway logo on dark glass",
)
import base64
open("out.png", "wb").write(base64.b64decode(img.data[0].b64_json))

Video generation (AI Studio)

Video is asynchronous: POST https://darkcoder.ai/v1/videos/generations returns a job id immediately (202); poll GET https://darkcoder.ai/v1/videos/generations/<id> until status is completed — then data[].url points at the stored asset. Higgsfield video models include hf/kling-3-turbo, hf/seedance-2, hf/wan-2-6 and hf/hailuo-2-3; pass image (a public image URL) for image-to-video.

curl -X POST https://darkcoder.ai/v1/videos/generations \
  -H "Authorization: Bearer dc_live_demo_cursor_home_0000000000003f9a" \
  -H "Content-Type: application/json" \
  -d '{"model":"hf/kling-3-turbo","prompt":"A slow dolly along a coast road","duration":5,"size":"1536x864"}'
# {"id":"st_…","status":"queued"}
curl https://darkcoder.ai/v1/videos/generations/st_… -H "Authorization: Bearer dc_live_demo_cursor_home_0000000000003f9a"

Account vs API key

Every provider declares which connect modes it supports. account mode signs in with the vendor's official flow — OAuth PKCE or device code, the same flows the vendor CLIs use — so your subscription (Copilot Pro, Claude Max, ChatGPT Plus/Pro, Google AI Pro) serves the gateway. apikey mode takes a pay-per-token key. Both can exist for one provider; you pick in the connect drawer.

Where the vendor supports it, account mode offers a Temporary chat switch — gateway conversations are ephemeral and never land in your account's chat history. DarkCoder.ai never reads browser cookies, never asks for the vendor password, and never replays web-chat sessions.

ProviderAccountAPI keyNote
GitHub Copilotdevice code (live)—VS Code client identity; see the warning in the drawer
Claudeblocked by Anthropic ToSAnthropic keyConsumer terms forbid Pro/Max OAuth tokens outside Claude Code
OpenAI · ChatGPTcoming nextOpenAI key
Geminicoming laterAI Studio key
DeepSeek—DeepSeek keyNo third-party sign-in exists
Groq—Groq keyFree tier also available
Kimi—Moonshot key
GLM—BigModel key
Cursor——No public API
Antigravity——No public API
Ollama——Local endpoint

Copilot's device flow is live today and uses GitHub's undocumented Copilot API through the VS Code client identity (as Aider, Continue and OpenCode do) — GitHub may consider this against Copilot terms. Claude account sign-in stays disabled: Anthropic's terms forbid Pro/Max OAuth tokens outside Claude Code, enforced since Jan 2026.

Mock vs real

In demo mode (DC_DEMO_MODE=true) every adapter is a mock so the cockpit runs offline. Outside demo mode only real adapters answer: GitHub Copilot via the device flow today, with API-key adapters for the key providers next. A provider whose adapter is not implemented yet fails loudly — nothing pretends to connect.