Docs   IntegrationsModel ProvidersOpenAI

OpenAI Integration

The OpenAI provider in Rig offers integration with OpenAI’s API services, supporting both completion and embedding models. It provides a client-based architecture for interacting with OpenAI’s models.

Key Features

  • Full support for GPT-3.5 and GPT-4 model families
  • Text embedding models (Ada, text-embedding-3)
  • Automatic token usage tracking
  • Tool/function calling support
  • Custom API endpoint configuration

Basic Usage

use rig::providers::openai;
 
// Create client from environment variable
let client = openai::Client::from_env();
 
// Or explicitly with API key
let client = openai::Client::new("your-api-key");
 
// Create a completion model
let gpt4 = client.completion_model(openai::GPT_4);
 
// Create an embedding model
let embedder = client.embedding_model(openai::TEXT_EMBEDDING_3_LARGE);

Available Models

Completion Models

  • GPT_4 / GPT_4O: GPT-4 base and optimized versions
  • GPT_35_TURBO: GPT-3.5 Turbo and its variants
  • GPT_35_TURBO_INSTRUCT: Instruction-tuned GPT-3.5

Embedding Models

  • TEXT_EMBEDDING_3_LARGE: 3072 dimensions
  • TEXT_EMBEDDING_3_SMALL: 1536 dimensions
  • TEXT_EMBEDDING_ADA_002: 1536 dimensions (legacy)

Special Considerations

  1. Tool Calling: OpenAI models support function calling through a specialized JSON format. The provider automatically handles conversion between Rig’s tool definitions and OpenAI’s expected format.

  2. Response Processing: The provider implements special handling for:

    • Tool/function call responses
    • System messages
    • Token usage tracking

  3. Error Handling: OpenAI-specific errors are automatically converted to Rig’s error types for consistent error handling across providers.

Implementation Details

The core OpenAI provider implementation can be found in:

rig-core/src/providers/openai.rs [1:150]
//! OpenAI API client and Rig integration
//!
//! # Example
//! ```
//! use rig::providers::openai;
//!
//! let client = openai::Client::new("YOUR_API_KEY");
//!
//! let gpt4o = client.completion_model(openai::GPT_4O);
//! ```
use crate::{
    agent::AgentBuilder,
    completion::{self, CompletionError, CompletionRequest},
    embeddings::{self, EmbeddingError, EmbeddingsBuilder},
    extractor::ExtractorBuilder,
    json_utils, Embed,
};
use schemars::JsonSchema;
use serde::{Deserialize, Serialize};
use serde_json::json;
 
// ================================================================
// Main OpenAI Client
// ================================================================
const OPENAI_API_BASE_URL: &str = "https://api.openai.com";
 
#[derive(Clone)]
pub struct Client {
    base_url: String,
    http_client: reqwest::Client,
}
 
impl Client {
    /// Create a new OpenAI client with the given API key.
    pub fn new(api_key: &str) -> Self {
        Self::from_url(api_key, OPENAI_API_BASE_URL)
    }
 
    /// Create a new OpenAI client with the given API key and base API URL.

Tool calling and response handling:

rig-core/src/providers/openai.rs [350:450]
#[derive(Debug, Deserialize)]
pub struct Choice {
    pub index: usize,
    pub message: Message,
    pub logprobs: Option<serde_json::Value>,
    pub finish_reason: String,
}
 
#[derive(Debug, Deserialize)]
pub struct Message {
    pub role: String,
    pub content: Option<String>,
    pub tool_calls: Option<Vec<ToolCall>>,
}
 
#[derive(Debug, Deserialize)]
pub struct ToolCall {
    pub id: String,
    pub r#type: String,
    pub function: Function,
}
 
#[derive(Clone, Debug, Deserialize, Serialize)]
pub struct ToolDefinition {
    pub r#type: String,
    pub function: completion::ToolDefinition,
}
 
impl From<completion::ToolDefinition> for ToolDefinition {
    fn from(tool: completion::ToolDefinition) -> Self {
        Self {
            r#type: "function".into(),
            function: tool,
        }
    }

For detailed API reference and additional features, see the OpenAI API documentation and Rig’s API documentation.


API Reference (docs.rs)
AnthropicVector Stores