Crate polaris_models 

Model provider interface and registry for Polaris.

Provides a unified interface for AI model access, decoupling consumers from provider implementations.

Overview

• Provider-agnostic: Consumers depend only on this crate, not specific provider plugins.

• Modular provider plugins: Provider crates register at runtime, allowing models to be swapped via configuration without code changes.

• Minimal dependencies: Each provider lives in a separate crate.

Example

use polaris_models::{ModelRegistry, ModelsPlugin};
use polaris_models::llm::LlmRequest;
use polaris_system::param::Res;
use polaris_system::system::SystemError;
use polaris_system::system;

#[system]
async fn my_agent(registry: Res<ModelRegistry>) -> Result<String, SystemError> {
    let llm = registry
        .llm("openai/gpt-4o")
        .map_err(|e| SystemError::ExecutionError(e.to_string()))?;

    let response = llm.builder()
        .system("You are helpful")
        .user("Hello!")
        .generate()
        .await
        .map_err(|e| SystemError::ExecutionError(e.to_string()))?;

    Ok(response.text())
}

See Also

For the full framework guide, architecture overview, and integration patterns, see the polaris-ai crate documentation.

Re-exports

pub use tokenizer::Tokenizer;

pub use tokenizer::TokenizerPlugin;

Modules

dashboarddashboard

Dashboard snapshot endpoint for the model provider registry.

error

Error types for the model registry.

llm

LLM (Large Language Model) generation capabilities.

tokenizer

Token counting for LLM requests and messages.

Structs

ModelRegistry

Registry for model provider implementations.

ModelsPlugin

Plugin that provides the ModelRegistry for provider-agnostic model access.