Base URL

https://carbide-ai.work.gd/v1

OpenAI-compatible. Point any OpenAI SDK at Carbide by changing the base URL and the key — everything else stays the same.

from openai import OpenAI

client = OpenAI(
    base_url="https://carbide-ai.work.gd/v1",
    api_key=os.environ["CARBIDE_KEY"],
)

Authentication

Sign in with Discord at /login — your first sign-in creates the account and a sk-carbide-… key. Send the key as a bearer token on every request:

Keys are shown once and stored hashed. Rotate any time from the dashboard.

Chat completions

curl https://carbide-ai.work.gd/v1/chat/completions \
  -H "Authorization: Bearer $CARBIDE_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dsv4-flash-bf16",
    "messages": [{"role": "user", "content": "hello"}]
  }'

Responses include an usage object with token counts. Carbide deducts those tokens from your balance and returns two extra headers: X-Carbide-Tokens-Used and X-Carbide-Balance.

response = client.chat.completions.create(
    model="dsv4-flash-bf16",
    messages=[{"role": "user", "content": "hello"}],
)
print(response.choices[0].message.content)

Streaming

Pass "stream": true to receive server-sent events, exactly like the OpenAI API. Streamed usage is metered with a per-token estimate.

stream = client.chat.completions.create(
    model="dsv4-flash-bf16",
    messages=[{"role": "user", "content": "count to five"}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

List models

Public endpoint — no key required. The table below is live from the catalogue.

Earning tokens

Every ad you watch in the dashboard credits 100,000 tokens to your balance. Watch time and a small proof-of-work are verified server-side, each ad can be claimed once, ad blockers must be off, and there's a daily cap (50 per account and 100 per network, resets at midnight UTC). Spend tokens by calling the API — balances can go slightly negative on a single large request.

• POST /api/ads/start — begin a verified ad session (returns a proof-of-work challenge)
• POST /api/ads/claim — credit the reward after watch time + PoW
• GET /api/account — balance, usage, and ad counters

Errors & limits

Errors follow the OpenAI shape: {"error": {"message", "type", "code"}}.

• 401 — missing or invalid API key
• 402 — insufficient_balance: watch an ad to top up
• 404 — unknown model (check GET /v1/models)
• 429 — rate limit: 20 requests/min per key, 40/min per IP, max 2 concurrent per key, 90/min on account endpoints

Request bodies are capped at 8 MB. Idle balance of 0 tokens blocks new requests until you earn more.