Struct Provider 

pub struct Provider { /* private fields */ }

One provider: a backend, its concurrency budget, and the two pieces of run-scoped state that belong to it.

The limiter is per provider rather than per process because the slot represents in-flight backend work, and max_concurrent is configured per [[llm]] entry. A single shared limiter would apply the local model’s generous budget to a rate-limited cloud endpoint.

down and served live here rather than in vectors parallel to the provider list. Parallel vectors force a length invariant the type system cannot check, and make every reader defensive about an index that is structurally always valid — a mark_down that silently did nothing on an out-of-range index would be invisible.

The cache identity is computed once because it includes stable backend metadata such as the Codex CLI version. The model and display location stay on the backend so reporting cannot drift from the client that is used.

Implementations

impl Provider

pub fn model(&self) -> &str

The model this provider asks for.

pub fn location(&self) -> &str

The backend-neutral location this provider talks to.

pub fn limiter(&self) -> &Limiter

The concurrency budget for this provider’s backend.

Exposed because it is the only thing that can demonstrate the limiter works: wiremock never overlaps requests, so neither an in-flight counter nor wall-clock can tell a working limiter from a deleted one. Observing available() while the analysis runs can.

pub fn served(&self) -> usize

How many files this provider answered.

Counted here rather than carried out through the analysis result and rejoined against the chain later: the chain already knows who answered, it is already shared for the whole process, and it already holds per-provider interior-mutable state. A cache hit counts — the answer still originated with that provider, and the user’s code was still reviewed by that model.

pub fn is_down(&self) -> bool

Whether this provider has been demoted for the rest of the run.

pub fn cache_key( &self, cache: &Cache, system_prompt: &str, user_content: &str, ) -> CacheKey

This provider’s cache key for one prompt.

The single definition of “which key belongs to which provider”. It lives here rather than at the call site so a test cannot compute the key a different way than production does - which is exactly how the endpoint went missing from it: the tests spelled out model and temperature by hand and agreed with the bug.

Trait Implementations

impl Debug for Provider

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

Formats the value using the given formatter.