Struct OpenAiModelConfig 

pub struct OpenAiModelConfig {
    pub model: String,
    pub base_url: String,
    pub backend: OpenAiBackend,
    pub bearer: String,
    pub max_tokens_cap: u32,
    pub reasoning_summary: Option<String>,
    pub prompt_cache_key: Option<String>,
    pub custom_tools_supported: bool,
    pub system_placement: SystemPlacement,
    pub extra_headers: Vec<(String, String)>,
    pub provider_prefs: Option<Value>,
}

The per-model config record for the OpenAI wire family (plan §3.2 + §A.5).

Fields

model: String

The model id (gpt-… native; openai/…/x-ai/… via OpenRouter).

base_url: String

The endpoint base URL (the wire appends its request path).

backend: OpenAiBackend

The endpoint family — quirks only (auth is always Bearer).

bearer: String

The bearer credential.

max_tokens_cap: u32

Clamp for max_output_tokens (which includes reasoning tokens).

reasoning_summary: Option<String>

reasoning.summary request value. Default Some("auto") (plan §A.5 Q2: trace readability); None omits the field for score-only runs.

prompt_cache_key: Option<String>

prompt_cache_key — the facade passes the session id (plan §A.5 Q4; codex’s behavior; probe P6 confirmed harmless for xAI models).

custom_tools_supported: bool

Whether the target model/provider accepts type:"custom" tools. Manual flag (plan §A.5 Q5 — no model-id sniffing): xAI 422s them (probe P3); false degrades freeform specs to the {"input": string} function framing on both the tools array and the call/replay items.

system_placement: SystemPlacement

Where Role::System goes: codex’s top-level instructions (default) or grok’s in-stream role:"system" message.

extra_headers: Vec<(String, String)>

Extra headers appended to every request.

provider_prefs: Option<Value>

OpenRouter routing preferences; None → the default pair (allow_fallbacks:false, require_parameters:true) on OpenRouter.

Implementations

impl OpenAiModelConfig

pub fn new( model: impl Into<String>, base_url: impl Into<String>, bearer: impl Into<String>, ) -> Self

Build a config, detecting the backend from base_url.

pub fn from_env() -> Result<Self, ProviderError>

Resolve from the environment: LOCODE_API_KEY (required), LOCODE_BASE_URL (default native). The model defaults to DEFAULT_OPENAI_MODEL (namespaced openai/… on OpenRouter) — select it via --model/settings (ADR-0024 §1.4), not env.

Errors

ProviderError::Auth when LOCODE_API_KEY is unset or empty.

pub fn effective_provider_prefs(&self) -> Option<Value>

The OpenRouter provider preferences to inject, if any: the configured value, or {allow_fallbacks:false} on the OpenRouter backend.

require_parameters is deliberately absent (live finding, 2026-07-19): on the beta /v1/responses endpoint, require_parameters: true combined with a tools array 404s with “No endpoints found that can handle the requested parameters” — OpenRouter’s parameter registry does not count tools as supported there. The silent-param-drop protection it provided on the Messages endpoint is covered here by the live smokes instead.

Trait Implementations

impl Clone for OpenAiModelConfig

fn clone(&self) -> OpenAiModelConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for OpenAiModelConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.