Module template

Abstractions that tie a prompt to a concrete model and a typed response.

The artificial framework purposely keeps the public surface small. A developer usually needs only two traits to go from “some string fragments” to “ready-to-send payload”:

1. IntoPrompt – turns any value into a list of chat messages.
2. PromptTemplate – adds metadata such as the target model and the expected JSON response schema.

Provider back-ends (e.g. artificial-openai) accept any P that implements both traits. Thanks to Rust’s powerful type system the compiler guarantees at compile time that

• the message type produced by the prompt matches what the back-end expects,
• the JSON returned by the provider can be deserialised into P::Output.

use artificial_core::template::{IntoPrompt, PromptTemplate};
use artificial_core::generic::{GenericMessage, GenericRole};
use artificial_core::model::{Model, OpenAiModel};
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};

#[derive(Serialize, Deserialize, JsonSchema)]
#[serde(deny_unknown_fields)]
struct Hello { greeting: String }

struct HelloPrompt;

impl IntoPrompt for HelloPrompt {
    type Message = GenericMessage;
    fn into_prompt(self) -> Vec<Self::Message> {
        vec![GenericMessage::new("Say hello!".into(), GenericRole::User)]
    }
}

impl PromptTemplate for HelloPrompt {
    type Output = Hello;
    const MODEL: Model = Model::OpenAi(OpenAiModel::Gpt4oMini);
}

See examples/openai_hello_world.rs for a fully working program.

Traits

IntoPrompt

Converts a value into a series of chat messages.

PromptTemplate

High-level description of a prompt.