Get started in 5 minutes.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.inferhub.dev/v1",
    api_key="sk-airo-...",
)

resp = client.chat.completions.create(
    model="ocg/glm-5.1",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)

from anthropic import Anthropic

# IMPORTANT: Anthropic clients append /v1 themselves.
# Use the API root, NOT ".../v1" (that becomes /v1/v1/messages).
client = Anthropic(
    base_url="https://api.inferhub.dev",
    api_key="sk-airo-...",
)
# Or skip manual env setup:
#   npx @inferhub/helper
# Claude Code:
#   ANTHROPIC_BASE_URL=https://api.inferhub.dev
#   ANTHROPIC_AUTH_TOKEN=sk-airo-...
#   ANTHROPIC_MODEL=free/grok/grok-4.5

resp = client.messages.create(
    model="zai/glm-4.6",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.content[0].text)

curl https://api.inferhub.dev/v1/chat/completions \
  -H "Authorization: Bearer sk-airo-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ocg/glm-5.1",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

curl https://api.inferhub.dev/v1/images/generations   -H "Authorization: Bearer sk-airo-..."   -H "Content-Type: application/json"   -d '{
    "model": "gpt-image-2",
    "prompt": "A single red apple on a plain white studio background.",
    "size": "1024x1024"
  }'

# The image comes back inline: data[0].b64_json. There is no hosted URL.
# quality (low/medium/high) changes what the image costs, a lot.
# Image models also answer on /v1/chat/completions, where the picture arrives
# as an image_url content part.
#
# Anything the OpenAI shape has no field for goes in custom_params, which is
# passed to the upstream verbatim — this is how you reach the per-model knobs
# on rails like leo/* that expose a different parameter set per model:
#   "custom_params": { "contrast": 3.5, "style_ids": ["..."] }

# 1. Submit. This returns a JOB, not a video: generation takes 90s to several
#    minutes, which is longer than any HTTP request should stay open.
curl https://api.inferhub.dev/v1/videos   -H "Authorization: Bearer sk-airo-..."   -H "Content-Type: application/json"   -d '{
    "model": "leo/veo-3.1-fast-generate-001",
    "prompt": "A red cube slowly rotating on a white table.",
    "seconds": "8",
    "custom_params": { "motion_has_audio": false }
  }'
# -> {"id":"video_...","object":"video","status":"queued","progress":0}

# 2. Poll until status is "completed" (or "failed").
curl https://api.inferhub.dev/v1/videos/video_...   -H "Authorization: Bearer sk-airo-..."

# 3. Download the mp4.
curl https://api.inferhub.dev/v1/videos/video_.../content   -H "Authorization: Bearer sk-airo-..." -o out.mp4

# A generation that fails is not billed. The charge lands when the job
# completes, and appears on the job object as usage.cost.

curl https://api.inferhub.dev/v1/messages \
  -H "x-api-key: sk-airo-..." \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zai/glm-4.6",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'