Struct OpenRouterChatClient 

pub struct OpenRouterChatClient { /* private fields */ }

Process-wide OpenRouter chat client. Holds the model name so that callers only thread the per-item prompt/schema/input through Self::complete.

Implementations

impl OpenRouterChatClient

pub fn new( api_key: SecretBox<String>, model: String, timeout_secs: u64, ) -> Result<Self, AppError>

Builds a chat client bound to model, applying timeout_secs as the total per-request budget (wired from --openrouter-timeout). A value of 0 falls back to DEFAULT_TIMEOUT_SECS so a missing or zero flag never degrades into reqwest`’s immediate-timeout behaviour.

pub fn model(&self) -> &str

Returns the model bound to this client.

pub async fn complete( &self, system_prompt: &str, input_text: &str, schema_str: &str, max_tokens: Option<u32>, ) -> Result<ChatCompletion, ChatError>

Runs a single structured-output completion, transparently growing max_tokens and re-issuing the request when the model truncates its output (GAP-SG-70).

schema_str is the JSON Schema (as a string) the model must honour under strict: true. When input_text is empty only the system message is sent. max_tokens seeds the first attempt; None lets the provider apply its own default.

Returns ChatCompletion on success or ChatError on failure; both carry finish_reason/token diagnostics when a response was decoded.

Errors

Returns ChatError when: the schema is invalid JSON; the HTTP request fails or exhausts retries; the provider returns a permanent error (401/400/404, or a structured error object in a 2xx body); the response carries no usable content; the content cannot be parsed as JSON even after repair; the parsed JSON is not an object; or the response is truncated (finish_reason: "length") after crate::constants::ENRICH_MAX_LENGTH_RETRIES max_tokens growth attempts are exhausted.