Factorial Compute

Run and chain vision, 3D and reasoning models through one API.

Find every box in a warehouse photo, measure each one in 3D, and have a model write up which ones are damaged - three kinds of model, one result, and no glue code between them.

• Typed answers. Masks, boxes, depth maps, point clouds, meshes, tracks, transcripts and text come back in the same shape whichever model produced them.
• Chain by id. Pass one run's id as the next run's input. Files stay on our side; nothing is downloaded and uploaded again between steps.
• The price first. Every model's price is public, and POST /v1/quote prices a run before it starts. max_cost_usd refuses anything dearer.
• Plain REST. Any language works. There is a Python client, an OpenAI-compatible endpoint for language models, and tools for coding agents - all optional.

Start here

• Quickstart - an API key and a first answer, with curl, Python or Node.
• Chaining models - one model's answer as the next one's input.
• Models and prices - what you can run, live from the API.
• API reference - every endpoint, from the OpenAPI spec.

What a call looks like

curl https://api.factorialcompute.com/v1/executions \
  -H "Authorization: Bearer $FACTORIAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "sam3-segment",
       "input": {"image": "med_...", "objects": [{"id": "box", "text": "cardboard box"}]}}'

{
  "id": "exec_...",
  "status": "succeeded",
  "model": "sam3-segment",
  "output": {
    "kind": "masks",
    "objects": [{"id": "box", "box": [380, 151, 541, 311], "score": 0.97, "media_id": "med_..."}],
    "view_url": "https://api.factorialcompute.com/view#..."
  },
  "cost": {"usd": 0.0075, "complete": true}
}

Factorial is in private beta: an API key comes from an invitation. Ask for one at factorialcompute.com.