Testing with mock & observe

Test AI-flavored code deterministically, and trace what's actually happening when it runs.

test / expect

test "greet returns correct format":
    expect greet("Ada") == "Hello, Ada"

Run every test block in a project with:

nekova test

mock think

Stub an AI response so a test doesn't depend on what a real (or even mock) provider happens to say:

test "summary uses the right tone":
    mock think as "Hello there"
    let result = think "say hi"
    expect result == "Hello there"

mock think as <value> is scoped to the enclosing test block — it doesn't leak into other tests or into the rest of the program.

observe

Structured tracing for debugging AI-native code — wrap any block to see what ran and how long it took:

observe "pipeline run":
    let x = think "hello"
    show x

👁  pipeline run
🧠 [MOCK] Hello! I am NEKOVA's built-in AI assistant.
   └─ completed in 0.9ms

Tagging an observe block

observe "pipeline run" with tags {user: user_id}:
    let summary = think "summarize this"

Tags travel with the trace output, useful once you have more than one thing being observed in the same run.

Why mock responses are tagged

Every mock AI response is prefixed [MOCK], so output from mock think or the default mock provider can never be confused with a real model's reply — including in test output, where that distinction matters most.