Struct OpenAIAdapter 

pub struct OpenAIAdapter { /* private fields */ }

OpenAI API adapter.

Supports structured output generation via:

• Strict JSON schema mode (response_format with type: “json_schema”)
• Function calling (for GPT-4 and GPT-3.5-turbo)
• JSON mode (response_format with type: “json_object”)
• JSON schema validation (via function parameters)

Example

use cognee_llm::adapters::OpenAIAdapter;
use cognee_llm::Llm;

let adapter = OpenAIAdapter::new(
    "gpt-4-turbo-preview",
    "sk-...",
    None, // Use default base URL
)?;

let result: MyStruct = adapter.create_structured_output(
    "Extract information from this text",
    "You are a helpful assistant",
    None,
).await?;

Implementations

impl OpenAIAdapter

pub const DEFAULT_BASE_URL: &'static str = "https://api.openai.com/v1"

Default OpenAI API base URL

pub const DEFAULT_STRUCTURED_OUTPUT_RETRIES: usize = 5

Default retry attempts for structured output parsing paths.

Python parity: instructor’s acreate_structured_output retries up to MAX_RETRIES = 5 times on a parse/validation failure. We match that count so transient malformed responses get the same number of repair chances before the cognify pipeline gives up.

pub const DEFAULT_NETWORK_RETRIES: usize = 3

Default retry attempts for transient network/server errors.

pub fn new( model: impl Into<String>, api_key: impl Into<String>, base_url: Option<String>, ) -> Result<OpenAIAdapter, LlmError>

Create a new OpenAI adapter.

Arguments

• model - Model identifier (e.g., “gpt-4”, “gpt-3.5-turbo”)
• api_key - OpenAI API key
• base_url - Optional custom base URL (defaults to OpenAI’s API)

Returns

A new OpenAI adapter instance

pub fn with_structured_output_retries(self, retries: u32) -> OpenAIAdapter

Configure retry attempts for structured output extraction.

Values lower than 1 are coerced to 1.

pub fn with_network_retries(self, retries: u32) -> OpenAIAdapter

Configure retry attempts for transient network and server errors (HTTP 429, 5xx).

Each retry uses exponential backoff starting at 1 s, doubling up to 30 s.

pub fn with_transcription_model(self, model: impl Into<String>) -> OpenAIAdapter

Configure the model used for audio transcription (default: "whisper-1").

Trait Implementations

impl Clone for OpenAIAdapter

fn clone(&self) -> OpenAIAdapter

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Llm for OpenAIAdapter

fn generate<'life0, 'async_trait>( &'life0 self, messages: Vec<Message>, options: Option<GenerationOptions>, ) -> Pin<Box<dyn Future<Output = Result<GenerationResponse, LlmError>> + Send + 'async_trait>>

where 'life0: 'async_trait, OpenAIAdapter: 'async_trait,

Generate text completion from messages.

fn create_structured_output_with_messages_raw<'life0, 'life1, 'async_trait>( &'life0 self, messages: Vec<Message>, json_schema: &'life1 Value, options: Option<GenerationOptions>, ) -> Pin<Box<dyn Future<Output = Result<Value, LlmError>> + Send + 'async_trait>>

where 'life0: 'async_trait, 'life1: 'async_trait, OpenAIAdapter: 'async_trait,

Generate structured output from messages (type-erased).

fn model(&self) -> &str

Get the model identifier.

fn supports_streaming(&self) -> bool

Check if the LLM supports streaming.

fn supports_function_calling(&self) -> bool

Check if the LLM supports function calling / tool use.

fn max_context_length(&self) -> u32

Get the maximum context length (in tokens) for this model.

fn transcribe_image<'life0, 'life1, 'life2, 'async_trait>( &'life0 self, image_bytes: &'life1 [u8], mime_type: &'life2 str, options: Option<GenerationOptions>, ) -> Pin<Box<dyn Future<Output = Result<String, LlmError>> + Send + 'async_trait>>

where 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait, OpenAIAdapter: 'async_trait,

Describe the contents of an image using vision capabilities.

fn supports_vision(&self) -> bool

Whether this adapter supports image transcription.

fn create_structured_output_raw<'life0, 'life1, 'life2, 'life3, 'async_trait>( &'life0 self, text_input: &'life1 str, system_prompt: &'life2 str, json_schema: &'life3 Value, options: Option<GenerationOptions>, ) -> Pin<Box<dyn Future<Output = Result<Value, LlmError>> + Send + 'async_trait>>

where 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait, 'life3: 'async_trait, Self: 'async_trait,

Generate structured output from text (type-erased).

impl Transcriber for OpenAIAdapter

fn transcribe_audio<'life0, 'life1, 'life2, 'life3, 'life4, 'async_trait>( &'life0 self, audio: &'life1 [u8], format: &'life2 str, language_hint: Option<&'life3 str>, prompt_hint: Option<&'life4 str>, ) -> Pin<Box<dyn Future<Output = Result<TranscriptionOutput, LlmError>> + Send + 'async_trait>>

where 'life0: 'async_trait, 'life1: 'async_trait, 'life2: 'async_trait, 'life3: 'async_trait, 'life4: 'async_trait, OpenAIAdapter: 'async_trait,

Transcribe audio bytes to text.

fn transcription_model(&self) -> &str

Return the name of the transcription model in use.