Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.tryvoss.dev/llms.txt

Use this file to discover all available pages before exploring further.

The language exists because AI workflows repeat the same control patterns.

​
Confidence

probable<T> represents a value with confidence metadata. 
let intent: probable<string> = classify(message)

if intent @ 0.80 {
  return intent.value
}

return "unknown"

​
Context budgets

ctx makes token windows explicit. 
ctx (budget: 3000 tokens) {
  let answer = ask(model, prompt)
  return answer
}

​
Budget fallback

within budget expresses controlled degradation. 
within budget(tokens: 4000, latency: 10s) {
  return deep_research(topic)
} fallback {
  return quick_summary(topic)
}

​
Semantic routing

match similar routes by meaning rather than keywords. 
match message {
  similar("refund request") -> route_refund(message)
  similar("technical support") -> route_support(message)
  _ -> route_generic(message)
}

​
Agents

Agent primitives express lifecycle directly. 
agent Researcher(topic: string) -> Finding {
  return search_and_summarize(topic)
}

let handles = topics.map(t => spawn Researcher(t))
let findings = gather(handles)

​
Memory, tools, and errors

Voss also includes memory primitives, @tool definitions, prompt classes, try/catch, and use imports.
The goal is AI workflow control, not Python syntax parity.