`std.llm`

The provider layer with the language machinery taken off.

import std.llm

llm.chat({ user: "Say hello" })
llm.chat({ system: "Answer in one word.", user: "Capital of France?" })
llm.chat({ user: "Summarize this", model: "gpt-4o-mini" })

One prompt in, a string out. No typed coercion, no contracts, no retry loop, no tool calls.

Fields

user is required. system sets the system prompt. model overrides the configured model for that call only - passed through rather than written to the process environment, so concurrent workers under pmap cannot clobber each other's choice.

Offline

With no provider configured it returns a deterministic mock string rather than failing, on the same principle as ai - the program still runs. A failing live call raises a catchable error.

Use `ai` instead, nearly always

ai is the same provider underneath, plus everything that makes model output usable:

	`ai`	`llm.chat`
Typed output	yes	no
Contracts	yes	no
Coercion retry	yes	no
Tools	yes	no
Sessions	yes	no
Streaming	yes	no
Caching	yes	no
Voting	yes	no

ai "..." is shorter and does more. The reason llm.chat exists is for the few cases where you want the provider and nothing else: a one-off probe, a wrapper of your own, or code that needs the raw string with no machinery in the way.

If you find yourself rebuilding retry and parsing around llm.chat, that is what ai already is.

std.llm

Fields

Offline

Use ai instead, nearly always

`std.llm`

Use `ai` instead, nearly always