Research · Decision models
Wise advice, for every decision.
Metis weighs every possible answer and tells you how sure it is.
A decision model answers a question with a fixed set of possible answers. It writes nothing: it reads your case once and returns a probability for every option.
Small, fast and light on energy
A chat model writes its answer token by token. Metis reads your case and questions once, and generates no tokens at all.
- 72 ms Three answers about one ticket Measured on the support-ticket example below
- 0.0047 Wh Energy for that request From the impact report on the response
- 32 Questions per case, in one pass Yes or no, one option from a set, or a level on a scale
A real example
One support ticket, three questions.
A customer writes in: since 09:00 every checkout on their web shop fails, and their status page shows the payment service as down. Metis returns the full picture, not only the winner.
Is a live service down?
- Yes 95%
Which team should take this ticket?
- Technical 93%
- Billing 6%
- Account 1%
- Sales 1%
How urgent is it?
- Immediately 96%
Three kinds of question
Ask what your process needs to know.
-
Is a rule met?
Is something true, does a request pass a policy, is a live service down?
-
Which one?
Which team, which category, which next step in the workflow?
-
How much?
How urgent, how risky, how severe?
The probabilities are the point
Confident? Act on it. Unsure? Pass it to a person.
When Metis is confident, your process acts on the answer automatically. When it is not, the case goes to a person. You decide where that line sits, question by question.
That fits the routine decisions that pile up in a queue: routing tickets, checking a request against a policy, sorting documents, choosing the next action. The clear cases move on their own.
How we use it ourselves
We built Metis because we needed it.
- 5× Faster routine decisions In our Mac app and in browser use
- In testing Our model router Deciding in one pass whether a smaller, more efficient model is enough
- EU Hosting Served from the EU endpoint only, on renewable energy. We don't train on your data.
What Metis does not do
Know its limits.
-
It does not look things up.
It applies the rules and facts you give it, and it is not legal, tax or investment advice.
-
It returns probabilities, not explanations.
When you want the reasoning behind a decision, ask a chat model.
-
It does not calculate reliably.
Put computed figures in the case, such as an order of €61,800 against a limit of €24,000.
Try Metis
One request, every answer.
Use your GreenPT API key, send a case with your questions, and set the model to metis.
curl https://api.greenpt.ai/v1/systemone \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer sk-your_api_key' \
--data '{
"model": "metis",
"state": "Support ticket #T-4821 from an enterprise customer: \"Since 09:00 every checkout on our web shop fails with error 502, so we cannot take any orders. Our status page shows the payment service as down. Please help, this is costing us sales.\"",
"questions": {
"outage": {
"type": "noul",
"instructions": "Is a live service down?",
"criteria": { "true": "a production service is unavailable", "false": "every service is reachable" }
},
"urgency": {
"type": "score",
"instructions": "How urgent is this ticket?",
"criteria": ["can wait", "this week", "today", "immediately"]
}
}
}'Decision models, in short
What is a decision model?
A decision model answers a question with a fixed set of possible answers. Instead of writing text, it returns a probability for every option, so you can see which answer it favours and how sure it is.
How is Metis different from a chat model?
A chat model writes its answer token by token. Metis writes nothing: it reads your case and questions once and returns a probability for each answer. That makes it faster and lighter on energy, and its answers plug straight into your own logic.
Where is Metis hosted?
Metis is served from GreenPT's EU endpoint only, on data centres that run on renewable energy. We don't train on your data.
How do I start with Metis?
Send a POST request to https://api.greenpt.ai/v1/systemone with the model metis, your case and your questions. The Metis guide in the GreenPT docs has a complete example.
Read the Metis guide →