Struct ResolvedEmbeddings 

pub struct ResolvedEmbeddings {
    pub backend: String,
    pub url: String,
    pub model: String,
    pub backfill_batch: u32,
    pub embedding_dim: Option<u32>,
    pub requested_dim: Option<u32>,
    pub key_source: KeySource,
    pub source: ConfigSource,
    /* private fields */
}

Canonical resolved-embedder configuration. Produced by AppConfig::resolve_embeddings.

Secret handling (#1598). The api_key field is private; access via Self::api_key. The manual Debug impl redacts the value (<redacted>), mirroring ResolvedLlm.

Fields

backend: String

Embedding backend selector. "ollama" (local /api/embed wire shape) or, since #1598, any #1067 OpenAI-compatible alias / the generic openai-compatible escape hatch. Classify via is_api_embed_backend.

url: String

Embedding endpoint base URL. The [embeddings].base_url / [embeddings].url synonym merge happens in the resolver (base_url wins); the field keeps the historical url name to limit call-site churn (#1598).

model: String

Embedding model identifier (canonicalised — legacy aliases nomic_embed_v15 / mini_lm_l6_v2 are mapped to the EmbeddingModel enum’s canonical HF id at resolve time).

backfill_batch: u32

Backfill batch size. Bounded 1..=10000; out-of-range values fall back to 100 with a WARN.

embedding_dim: Option<u32>

v0.7.x (issue #1169) — vector dim of the resolved model, when known. #1598: the explicit [embeddings].dim override wins over the canonical_embedding_dim table lookup. None when the operator chose a model id that isn’t in the table and set no override — in that case build_capability_models falls back to the tier preset’s dim (preserving pre-#1169 behaviour for unrecognised ids and avoiding the silent-wrong-dim trap for the recognised ones).

requested_dim: Option<u32>

#1598 (fleet follow-up) — the EXPLICIT [embeddings].dim override only (never table-derived). For OpenAI-compatible backends this is also sent as the wire dimensions request param, so Matryoshka-capable API models (gemini-embedding-2, text-embedding-3-*) return truncated vectors at the operator’s declared dim — the mechanism that keeps pgvector vector(768) fleet schemas + ANN indexes (≤2000-dim limit) usable with high-dim API models. None = model-native dim.

key_source: KeySource

#1598 — provenance of the resolved API key for boot-banner / doctor-probe display.

source: ConfigSource

Provenance of the resolved configuration.

Implementations

impl ResolvedEmbeddings

pub fn api_key(&self) -> Option<&str>

Access the resolved embedding API key. Use this only when constructing the embed client; do NOT log or {:?} the result.

pub fn from_parts( backend: String, url: String, model: String, embedding_dim: Option<u32>, api_key: Option<String>, ) -> Self

#1598 — construct from explicit parts. Prefer AppConfig::resolve_embeddings; this exists for tests and sibling surfaces (e.g. the reembed CLI) that synthesise a resolved view without an AppConfig.

pub fn with_requested_dim(self, dim: Option<u32>) -> Self

#1598 (fleet follow-up) — builder for the explicit requested output dimensionality (see Self::requested_dim).

Trait Implementations

impl Clone for ResolvedEmbeddings

fn clone(&self) -> ResolvedEmbeddings

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ResolvedEmbeddings

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

Formats the value using the given formatter.

impl Eq for ResolvedEmbeddings

impl PartialEq for ResolvedEmbeddings

fn eq(&self, other: &ResolvedEmbeddings) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for ResolvedEmbeddings