Struct OpenAiCompatAdapter 

pub struct OpenAiCompatAdapter { /* private fields */ }

HTTP adapter for any OpenAI-compatible /v1/chat/completions server.

Construct via OpenAiCompatAdapter::new. The adapter is Send + Sync and may be held behind an Arc<dyn LlmAdapter>.

Implementations

impl OpenAiCompatAdapter

pub fn new( base_url: impl Into<String>, model: impl Into<String>, api_key: Option<String>, timeout_ms: u64, max_sensitivity: Option<MaxSensitivity>, ) -> Result<Self, LlmError>

Construct an adapter.

api_key accepts None or an empty-string Some(""); both result in no Authorization header being sent.

max_sensitivity controls the data-classification gate before remote dispatch. Pass None to use the default of MaxSensitivity::Medium, which blocks high-sensitivity memories from being sent to the endpoint.

A warning is printed to stderr when base_url resolves to a non-loopback host, because all prompt content will be sent to that remote server.

Returns LlmError::InvalidRequest when:

• base_url does not start with http:// or https://.
• base_url contains no host.
• model is empty.

pub fn runtime_ceiling(&self) -> RuntimeCeiling

Return the runtime ceiling determined at construction from base_url.

Trait Implementations

impl Clone for OpenAiCompatAdapter

fn clone(&self) -> OpenAiCompatAdapter

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for OpenAiCompatAdapter

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

Formats the value using the given formatter.

impl LlmAdapter for OpenAiCompatAdapter

fn stream_boxed(&self, req: LlmRequest) -> BoxStream<'_>

Override with true OpenAI SSE streaming via POST /v1/chat/completions with "stream": true.

Parses data: lines, extracts choices[0].delta.content, and skips the terminal data: [DONE] sentinel.

fn adapter_id(&self) -> &'static str

Stable, lowercase identifier used in audit envelopes (e.g. "claude", "ollama", "replay"). Constants — implementations MUST NOT vary this per call.

fn complete<'life0, 'async_trait>( &'life0 self, req: LlmRequest, ) -> Pin<Box<dyn Future<Output = Result<LlmResponse, LlmError>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait,

Issue a completion call. The adapter is responsible for honouring req.timeout_ms and returning LlmError::Timeout when exceeded.

fn stream( &self, req: LlmRequest, ) -> impl Stream<Item = Result<StreamChunk, LlmError>> + Send

where Self: Sized,

Stream tokens incrementally. Default implementation calls complete and yields the full response as a single chunk — adapters override this for true streaming.