Learning

AI answers a situation while it is new. Owctopus stores the run, learns a condition from the repeats, and uses that condition the next time so it can skip the model.

What gets stored

Each run keeps the pieces Owctopus needs in order to learn:

• The context you sent, such as cart value, calories so far, or sleep hours.
• The options you sent, with their ids and fields.
• The decision, the reason, and the confidence.
• The outcome, once your app reports what happened.

From repeats to a condition

One AI answer does not become a rule. Owctopus waits until the same kind of situation shows up often, and the choice keeps working. That pattern is proposed as a candidate. A candidate is tested against past decisions and outcomes. After it is approved, it is active.

A shipping candidate can look like this:

{
  "condition": {
    "customerType": "premium",
    "urgency": "high",
    "cartValue": { "operator": ">", "value": 3000 }
  },
  "decision": "express",
  "confidence": 0.96,
  "sampleSize": 1840,
  "status": "candidate"
}

Status stays candidate until the condition is tested and approved. Only an active condition answers a later request on its own.

The same loop in other apps

A food logger can learn that a high-hunger lunch, with calories still well under target, should be the higher-protein bowl. A training app can learn that acceptable sleep, a week behind target, and 30 free minutes should be a short strength session.

Those conditions are created from the context and options your app keeps sending. They live on the Owctopus API, so TypeScript, Python, Go, and a plain HTTP call all receive the same learned decision. The next time the situation matches, the decision comes back from the condition, and AI is reserved for what is still unknown.

The dates describe an application whose decisions repeat. A one-off situation stays with AI.