think

The keyword that calls an AI model — no client, no SDK, no response object to unwrap.

Basic usage

let summary = think "Summarize the plot of Romeo and Juliet in one sentence"
show summary

think sends the prompt to the configured AI provider and returns the response directly.

The mock provider

By default, NEKOVA uses a mock provider so you can write and test AI-flavored code without an API key. Every mock response is tagged [MOCK], so it's never mistaken for a real model reply — this was a deliberate fix (see the Changelog) after early versions had some untagged mock branches.

Output formats

think can ask for a specific shape back:

let data = think "Extract fields from this text" as json
let items = think "List three colors" as list
let is_valid = think "Is this a valid email?" as bool
let count = think "How many legs does a spider have?" as number

Structured schemas

let user = think "Extract name and age from: Ada, 30" as schema {"name": "text", "age": "number"}

Inline error handling

If the AI call fails, when error: lets you supply a fallback instead of embedding an error string in the result:

let summary = think "summarize this document" when error: "unavailable"

Without the when error: clause, a failed call embeds [think error: ...] in the result rather than crashing the program — when error: exists for when you'd rather control that fallback yourself.

let x = think "hi" as json when error: {"status": "down"}

f-strings in prompts

task summarize(document):
    return think f"Summarize this in one sentence: {document}"

Timeouts

think calls run with a timeout under the hood, so a hung provider doesn't hang your whole program — see async & await for how NEKOVA handles this without you needing to manage an event loop yourself.