Write standard deterministic programs with primitives, functions, modules, and storage. When your project benefits from model-assisted reasoning, invoke it directly as a first-class language primitive.
General-purpose programming; AI remains optional
Opt-in AI primitives with strict guardrails
prompt generateCode { "Write a function to reverse a string"}let code = model("gemini-3.1-pro-preview") { generateCode }show codemodel("local") for local inference, or configure supported external providers with environment credentials.
Reasoning does not replace Sesi's regular control flow. Define the task, choose where it runs, then work with the result alongside the rest of your program.
Use a named prompt block to combine instructions with values from your program.
Run locally or select a configured provider at the point where reasoning is needed.
Assign the response to a value, display it, store it, or pass it into the next part of your program.
let topic = "event loops"prompt explain {"Explain " topic " in three short points."}let answer = model("local") { explain }show answermodel("local")
Open documentation
Keep exact work in ordinary Sesi. Reach for a model when the job depends on interpretation, synthesis, classification, or natural-language output.
Use functions, conditions, loops, modules, storage, and network tools when the program should produce an exact, repeatable result.
Use a prompt and model when the program needs to understand open-ended input or produce an answer that cannot be expressed as fixed rules.
Nothing calls a model unless the program reaches a model expression.
Select model("local") to keep inference on your machine.
Prompts and model calls live beside ordinary variables and functions instead of in a separate framework.
Explore the core syntax, then return here when your program needs model-assisted work.