Build with AI primitives

The parts a builder starts with: models, APIs and agents, each described factually. The snippets are sketches, not code for any real service.

01 A model

A trained system that maps an input to an output. You choose it by task, size and cost.

// sketch only, not a real API
input  = "Summarize this page"
output = model.run(input)

The network door to a model. It takes a request, returns a reply and has limits you read first.

// sketch only
reply = post(endpoint, { input, options })
check(reply.status)

A function the model may ask your program to run: search, a calculator, a database read.

// sketch only
tools = [ search, calculate ]
model.run(input, tools)

A model, some tools and a loop: act, look at the result, decide the next step, stop on a rule.

// sketch only
while not done and steps < limit:
    step = model.next(state)
    state = run(step)

A fixed set of inputs with expected outputs, run on every change, so you notice a regression.

// sketch only
for case in cases: assert grade(case)

Each primitive page would link the provider's own documentation.

$ echo "This guide is a design. The domain behind it is for sale."