Give your agent
a next move.

Bring text, one image or camera frame, or a 2–4-frame sampled video or camera window. Define two to eight allowed actions. Get the selected action and full probability distribution with zero generated text tokens.

Try the gesture demo Read the API docs

100 free Playground decisions · No card required · Then $0.042 / million billed input tokens

Start with an image or your camera in the Playground, then create an API key when you are ready to connect your agent.

01 / A GUI agent, right nowInput example

▣Current screenReport details

↗Visible controlExport menu

Goal: export the report as CSV

UI state

What should the agent do next?

01Open the Export menu
02Open report filters
03Wait for more information

Illustrative structured UI state, not a screenshot. Change the state to inspect a request; no model has been called here.

{
  "model": "trio-spark-v1.1",
  "task": "The user asked to export the current report as CSV. Choose the next safe UI action from the available controls.",
  "state": {
    "app": "Reports",
    "screen": "Report details",
    "visible_controls": [
      "Export menu",
      "Filters",
      "Help"
    ]
  },
  "choices": [
    {
      "id": "open_export_menu",
      "description": "Open the Export menu"
    },
    {
      "id": "open_filters",
      "description": "Open report filters"
    },
    {
      "id": "wait",
      "description": "Wait for more information"
    }
  ]
}

Try this request in the live Playground

01 / Your world

Bring the current state.

Use text or JSON, one JPEG or PNG image or camera frame, or a 2–4-frame sampled video or camera window. You define the task.

02 / Your choices

Define the next moves.

Supply 2–8 choices with your own IDs and descriptions. Keep the actions meaningful to your application.

03 / Your application

Make the decision useful.

Receive a choice and probabilities. Your code decides when to act, ask for review, or wait.

One request. Your use case.

Small decisions.
Room to build.

Start with a task you can check: selecting a GUI agent’s next control, choosing a Falling Blocks move, or routing a request. Try your own states and choices in the live Playground.

The API returns a choice ID, probabilities, a decision status, and usage. Review uncertain choices before turning them into actions.

Trio-Spark v1.1 accepts text, one image, or 2–4 sampled frames with timestamps. Video input represents that sampled window rather than an arbitrary video URL or continuous stream.

The same decision contract powers the Playground, API, and game demos.

curl https://platform.machinefi.com/api/spark/v1/decisions \
  -H "Authorization: Bearer $TRIO_SPARK_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
  "model": "trio-spark-v1.1",
  "task": "The user asked to export the current report as CSV. Choose the next safe UI action from the available controls.",
  "state": {
    "app": "Reports",
    "screen": "Report details",
    "visible_controls": [
      "Export menu",
      "Filters",
      "Help"
    ]
  },
  "choices": [
    {
      "id": "open_export_menu",
      "description": "Open the Export menu"
    },
    {
      "id": "open_filters",
      "description": "Open report filters"
    },
    {
      "id": "wait",
      "description": "Wait for more information"
    }
  ]
}'

Pay as you go

$0.042

per million billed input tokens for text and visual input · output free

$5minimum prepaid balance

No subscriptionpay for completed judgments

100 free decisions with a new account. Pay as you go after that, with top-ups starting at $5.

Decision intelligence
for real agent loops.

Trio-Spark v1.1 brings visual and text decisions from the Situated World Models program to developers through one hosted API. Start with a small, testable problem.

Test Spark on your own states and action sets before choosing an automation threshold.

Give your agenta next move.