Struct OpenAiCompatEmbedder 

pub struct OpenAiCompatEmbedder { /* private fields */ }

An /v1/embeddings client for any OpenAI-compatible server.

Implementations

impl OpenAiCompatEmbedder

pub fn new(endpoint_url: &str, model: &str, dim: usize) -> Self

Creates a client for the exact embeddings endpoint_url (e.g. https://api.openai.com/v1/embeddings or http://localhost:11434/v1/embeddings), a model name and the expected dimension. The model name is also the default semantic-space id; use Self::with_space_id to declare a stable revision or digest instead. The URL is used as supplied; this constructor does not append or otherwise rewrite a path. The dimension and identity are explicit — no startup probe request, and a server disagreeing with the dimension is a typed error, not a silently reconfigured database.

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

Adds a bearer API key (OpenAI et al.; local servers usually need none).

pub fn with_space_id(self, space_id: impl Into<String>) -> Self

Overrides the semantic vector-space identity persisted in the database. By default this is the model name passed to Self::new.

Use an explicit id when the provider’s request model is an alias, or when two differently named endpoints are known to produce compatible vectors. Plugmem trusts this declaration and never probes the provider to infer it. Invalid ids are rejected when the database first uses the embedder.

Trait Implementations

impl Debug for OpenAiCompatEmbedder

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

Formats the value using the given formatter.

impl Embedder for OpenAiCompatEmbedder

fn space_id(&self) -> &str

Stable, human-readable identity of the semantic vector space. Different models (or incompatible revisions of one model) must return different ids even when their dimensions match.

fn dim(&self) -> usize

Vector dimension this embedder produces. 0 disables the vector layer (the engine is fully functional without it).

fn embed(&self, texts: &[&str]) -> Result<Vec<Vec<f32>>, HostError>

Embeds every text, one vector per input, in input order.