Decision models.
One hosted API.

Try open decision models via simple API call. Decider handles the hosting and pays for the GPU.

No sign up
No credit card
No API key

Try the free API Contact me on X

Best-effort availability. Contact me about self-hosting on your own compute or the serving layer behind this API.

“Our payment service has been down for three hours.”

Is this urgent?noul

Yes0.68

Which team should handle this?choice

Engineering(picked)0.91
Billing0.08
Sales0.01

Sample answers. Two questions, one request.

Call it from your terminal

Choose Kev or Strands, send context and questions, and get probabilities, choices, or scores.

Model

Question type

POST/v1/systemone

curl https://api.decider.cloud/v1/systemone \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "kev-4b",
    "state": "Our payment service has been down for three hours.",
    "questions": {
      "urgent": {
        "type": "noul",
        "instructions": "Is this urgent?"
      }
    }
  }'

200Sample response

{
  "model": "kev-4b",
  "answers": {
    "urgent": {
      "type": "noul",
      "noul": 0.68235
    }
  },
  "usage": {
    "input_tokens": 23,
    "output_tokens": 19
  },
  "latency_ms": 143.59
}

Cold starts: The API scales to zero when idle. Your first request after inactivity can take about 30 seconds while models load.

Available decision models

Kev

kev-4b

86ms

Warm p50 · L40S GPU¹

Base model: Qwen3.5 4B
Context per question: 12,288 tokens
Question types: noul, choice, score²

Kev on GitHub

Strands Decider

strands-decider-2b

45ms

Warm p50 · L40S GPU¹

Base model: Qwen3.5 2B
Context per question: 4,096 tokens
Question types: noul, choice, score²

Strands Decider on GitHub

Latency. Median of sample warm requests. Includes input processing and inference; excludes network time and cold starts.

Shared limits. Both models take up to 8 questions per request and 255 options per choice.

Why decision models

Many problems need structured decisions, not generated prose or code. Decision models can return structured answers in real time, at low inference cost, without task-specific training.

Scroll to compare all approaches

Rule engines, ML classifiers, LLMs, and decision models
Requirement	Rule engine	ML classifier	LLM	Decision models
InputNatural language context and questions	×	△	✓	✓
OutputBounded choices and confidence	✓	✓	△	✓
GeneralizationPrompt it; no task training	×	×	✓	✓
LatencyFast decisions in the request path	✓	✓	△	✓
CostLow inference cost at scale	✓	✓	△	✓

Typical fit
Depends on implementation
Not a typical fit

About Decider

Why Decider?

Decider makes open decision models easier to evaluate through a shared API. Try different models on your own examples before choosing one. Kev and Strands are available today.

Is Decider free, and do I need an API key?

Yes, during the public preview. No API key is required. I set up the serving infra and pay for the GPU so you don't have to. The preview has a fixed compute budget and can change or end. Availability is best effort; use it for research and evaluation.

Can I use my TypeSafe request format?

Decider follows the TypeSafe SystemOne request and answer shapes. Change the endpoint and select a supported model. See the compatibility notes for details.

Why can the first request take longer?

The service scales to zero while idle. Loading the models can take about 30 seconds. The warm model medians above exclude that startup and network time.

What happens when the API is busy?

One inference runs at a time across all users and both models. A busy request returns 429 with Retry-After. Wait and retry. The public preview has no uptime guarantee.

Is this suitable for private data?

The serving runtime never stores request content. Decider logs only model names and request timing. Use the preview for evaluation with nonconfidential inputs.

Understanding decision models

What is Jev?

Jev is TypeSafe AI’s first public System One model for structured decisions. Decider hosts Kev and Strands independently. Read TypeSafe’s introduction.

What is an AI decision model?

It turns context and a question into a probability, choice, or score your software can use. Define the task and allowed answers in your request.

What is a System One model?

System One models make fast, structured decisions. System Two models use slower, deliberate reasoning. Learn about TypeSafe’s System One models.

How do decision models differ from LLMs?

LLMs generate text and can return structured data. Decision models focus on probabilities, choices, and scores. Use an LLM when you need writing or explanations.

What are decision models useful for?

Route tickets, classify content, prioritize incidents, or choose a workflow’s next step. They suit tasks with clear options or scoring criteria.

Can decision models make mistakes?

Yes. Structured answers can still be wrong. Test on your own data, set suitable thresholds, and use human review or fallbacks for consequential decisions.

Find out what decision models can do.

Bring your own examples. Try the API free, no API key needed.

Try the free API

Need your own compute or guaranteed capacity? Message me on X about a private deployment.

Decision models. One hosted API.