AI Providers

Alef treats AI provider calls as a standard-library surface.

Deterministic Stub

Use stub in docs and tests:

fn main() {
    let r = std.ai.response_with("stub", "Summarize this ticket")
    println(r.provider)
    println(r.text)
}

The stub provider makes examples runnable without credentials.

Configured Provider

let r = std.ai.response("Summarize this ticket")

Use environment variables to configure live providers.

Named Providers

let r = std.ai.response_with("ollama", "hello")
let r = std.ai.response_with("minimax", "hello")
let r = std.ai.response_with("mimo", "hello")

Provider availability is environment-gated. A skipped live provider check is

not a failure unless the task explicitly requires that provider.

call_llm and stream_llm

In addition to response and response_with, Alef supports lower-level interactions:

// Single call
let result = std.ai.call_llm("ollama", "Summarize this ticket");
println(result);

// Streaming call
std.ai.stream_llm("ollama", "Write a long story", fn(chunk) {
    println("Received chunk: " + chunk);
});

embed

Alef provides an embed function to generate embeddings from text.

let embedding = std.ai.embed("ollama", "Hello world");
println(embedding[0]); // First float in the embedding vector

Chat helpers

Higher-level chat APIs accept message lists (role + content). Prefer them for

multi-turn agents:

let r = std.ai.chat_with("stub", [
    { "role" => "system", "content" => "You are a concise assistant." },
    { "role" => "user", "content" => "Summarize: deploy failed on auth" }
])
println(r.text)

Streaming variants exist for token/chunk delivery (stream, chat_stream, and

*_with / *_options forms depending on release). Always keep a stub path

for docs and tests.

Tool-calling (advanced)

Some providers support tool/function calls via chat_with_tools style APIs.

Treat tool schemas as application data (maps/JSON), not as a separate language

feature. Gate live providers with environment checks.

Pairing AI with HTTP and jobs

A common production shape is:

1. std.http serves /api/chat, /api/history, health, and shutdown.

2. std.ai answers prompts (stub in CI, real provider in prod).

3. std.db stores sessions.

4. std.jobs runs background work with retry.

See What's New — July 2026 and the

AI workflow tutorial.