Analyze and outcomes

analyze asks a configured model for a value with the shape you declare. The call returns an outcome, so your program handles success, uncertainty, budget exhaustion, and provider failure explicitly.

Define the result type

type Expense:
    merchant: str
    amount: float
    category: str
    policy_violation: bool

The fields become the schema requested from the model. Keep the type small and give each field a clear name. Use the type annotation on the left of the call:

expense: Expense = analyze(
    "Steakhouse dinner, $680, client entertainment",
    "Extract the expense and flag meals over $200"
)

Handle all outcomes

match expense:
    case Ok(value):
        print(f"{value.merchant}: ${value.amount}")
    case Uncertain(reason):
        print(f"The model needs review: {reason}")
    case Exhausted(meter):
        print(f"Budget used up: {meter}")
    case Failed(why):
        print(f"Provider unavailable: {why}")

Uncertain means the model answered that it could not decide. Failed means the provider or a tool did not answer. They are different recovery paths. Exhausted means a Kora budget was reached; it is not a crash.

Use a specific model

summary: str = analyze(issue, "Summarize in two sentences", model="cheap")

The model name comes from [models] in kora.toml. See Model providers for configuration and replay.