Struct OpenAiClientBuilder 

pub struct OpenAiClientBuilder { /* private fields */ }

Builder for OpenAiClient.

Created via OpenAiClientBuilder::default or OpenAiClient::builder. All fields have sensible defaults except api_key, which must be set before build.

Implementations

impl OpenAiClientBuilder

pub fn with_api_key(self, key: impl Into<String>) -> Self

Set the API key for authentication.

Required — build returns an error if this is not set. The key is sent as the Authorization: Bearer <key> header on every request.

pub fn with_base_url(self, url: impl Into<String>) -> Self

Set the base URL for API requests.

Defaults to https://api.openai.com/v1. Override when targeting an OpenAI-compatible endpoint (e.g. https://api.deepseek.com/v1, http://localhost:11434/v1 for Ollama, or a vLLM server).

pub fn with_model(self, model: impl Into<String>) -> Self

Set the default model identifier.

The model string is sent as the model field on every request. Can be changed at runtime via OpenAiClient::set_model (e.g. when the FallbackManager trips).

pub fn with_timeout(self, timeout: Duration) -> Self

Set the total request timeout (connect + response + body).

Defaults to 120 seconds. Ignored when a client was supplied via with_http_client.

pub fn with_connect_timeout(self, timeout: Duration) -> Self

Set the TCP connection establishment timeout.

Defaults to 10 seconds. Ignored when a client was supplied via with_http_client.

pub fn with_http_client(self, client: Client) -> Self

Inject a pre-built, shared reqwest::Client.

When set, the client’s connection pool is shared with every other provider built from the same handle, and the pool/TCP knobs are ignored. Configure timeouts on the injected client, not here.

pub fn with_stream_usage(self, enabled: bool) -> Self

Control whether streaming requests include stream_options.include_usage.

Defaults to true — real OpenAI, DeepSeek, and Grok support it and send a final usage chunk. Pass false for OpenAI-compatible servers that reject the parameter with a validation error (older Ollama, some self-hosted vLLM/LM Studio deployments). When disabled, streamed turns report usage: None instead of real token counts.

Ignored on non-streaming requests.

pub fn with_pool_max_idle_per_host(self, n: usize) -> Self

Set the maximum idle connections kept alive per host.

Defaults to reqwest’s built-in default (unlimited). Ignored when a client was supplied via with_http_client.

pub fn with_pool_idle_timeout(self, d: Duration) -> Self

Set how long an idle connection stays in the pool before being closed.

Defaults to reqwest’s built-in default (90s). Ignored when a client was supplied via with_http_client.

pub fn with_tcp_keepalive(self, d: Duration) -> Self

Set the OS-level TCP keepalive interval.

Defaults to disabled (reqwest default). Ignored when a client was supplied via with_http_client.

pub fn with_tcp_nodelay(self, enabled: bool) -> Self

Control whether TCP_NODELAY is set on connections.

Defaults to true — SSE streaming benefits from disabling Nagle’s algorithm. Pass false to re-enable it. Ignored when a client was supplied via with_http_client.

pub fn build(self) -> Result<OpenAiClient, ApiError>

Build the client.

Errors

Returns ApiError if no API key was set.

Trait Implementations

impl Default for OpenAiClientBuilder

fn default() -> Self

Returns the “default value” for a type.