atlascloud

Atlas Cloud

Atlas Cloud provides chat, image, video, and media generation APIs behind one API key. The chat API is OpenAI-compatible; image and video jobs are asynchronous and return a prediction ID that must be polled.

Official docs: https://www.atlascloud.ai/docs/en/models/get-start

---

When to Use

Use this skill when you need to:

• Run OpenAI-compatible chat completions through Atlas Cloud
• Generate images from text prompts
• Generate videos from text prompts or uploaded images
• Upload media for image-to-video, image editing, or other model workflows
• Poll asynchronous generation jobs until they complete

---

Prerequisites

Connect the Atlas Cloud connector at app.vm0.ai/connectors.

Troubleshooting: If requests fail, run zero doctor check-connector --env-name ATLASCLAUDE_TOKEN or zero doctor check-connector --url https://api.atlascloud.ai/v1/chat/completions --method POST

---

How to Use

Atlas Cloud uses Authorization: Bearer $ATLASCLAUDE_TOKEN.

1. Chat Completion

Write to /tmp/atlascloud_chat.json:

{
  "model": "deepseek-v3",
  "messages": [
    { "role": "system", "content": "You are a concise assistant." },
    { "role": "user", "content": "Explain async image generation in one paragraph." }
  ]
}

Then run:

curl -s -X POST "https://api.atlascloud.ai/v1/chat/completions" --header "Authorization: Bearer $ATLASCLAUDE_TOKEN" --header "Content-Type: application/json" -d @/tmp/atlascloud_chat.json | jq -r '.choices[0].message.content'

2. Streaming Chat Completion

Write to /tmp/atlascloud_stream.json:

{
  "model": "deepseek-v3",
  "stream": true,
  "messages": [{ "role": "user", "content": "Count from 1 to 5." }]
}

Then run:

curl -sN -X POST "https://api.atlascloud.ai/v1/chat/completions" --header "Authorization: Bearer $ATLASCLAUDE_TOKEN" --header "Content-Type: application/json" -d @/tmp/atlascloud_stream.json

3. Generate an Image

Write to /tmp/atlascloud_image.json:

{
  "model": "seedream-3.0",
  "prompt": "A serene Japanese garden with cherry blossoms, watercolor style"
}

Then run:

curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateImage" --header "Authorization: Bearer $ATLASCLAUDE_TOKEN" --header "Content-Type: application/json" -d @/tmp/atlascloud_image.json | jq '{id: .data.id, status: .data.status}'

Save .data.id as the prediction ID.

4. Generate a Video

Write to /tmp/atlascloud_video.json:

{
  "model": "kling-v2.0",
  "prompt": "A rocket launching into space with dramatic lighting and smoke effects"
}

Then run:

curl -s -X POST "https://api.atlascloud.ai/api/v1/model/generateVideo" --header "Authorization: Bearer $ATLASCLAUDE_TOKEN" --header "Content-Type: application/json" -d @/tmp/atlascloud_video.json | jq '{id: .data.id, status: .data.status}'

5. Poll a Prediction

Replace <prediction-id> with the ID returned by an image or video request:

curl -s "https://api.atlascloud.ai/api/v1/model/prediction/<prediction-id>" --header "Authorization: Bearer $ATLASCLAUDE_TOKEN" | jq '{status: .data.status, outputs: .data.outputs, error: .data.error}'

Poll every 2-5 seconds until status is completed or failed.

6. Upload Media

Use uploaded media URLs for image-to-video or image-editing models:

curl -s -X POST "https://api.atlascloud.ai/api/v1/model/uploadMedia" --header "Authorization: Bearer $ATLASCLAUDE_TOKEN" -F "file=@/tmp/source.png" | jq -r '.url'

Then pass the returned URL to the model-specific request, usually as image_url.

---

Guidelines

1. Use the correct base URL: chat completions use https://api.atlascloud.ai/v1; image, video, upload, and polling APIs use https://api.atlascloud.ai/api/v1.
2. Treat image and video jobs as async: always save .data.id and poll before using outputs.
3. Check model pages before adding parameters: model-specific names and parameters vary; use the Atlas Cloud model library when a request needs sizing, duration, aspect ratio, seed, LoRA, or image input options.
4. Use short polling intervals for images and longer intervals for video: 2 seconds is reasonable for images; 5 seconds is better for videos.
5. Handle failures explicitly: if status is failed, read .data.error and adjust model name, parameters, input media URL, prompt, or account balance.