Module embedding 

Embedding provider abstraction and implementations.

Defines the EmbeddingProvider trait and concrete implementations:

• DisabledProvider — returns errors; used when embeddings are not configured.
• OpenAIProvider — calls the OpenAI embeddings API with batching, retry, and backoff.
• OllamaProvider — calls a local Ollama instance’s /api/embed endpoint.
• LocalProvider — runs models locally via fastembed (primary) or tract (musl/Intel Mac); no network calls after model download.

Also provides vector utilities for working with sqlite-vec:

• cosine_similarity — compute similarity between two embedding vectors
• vec_to_blob — encode a Vec<f32> as little-endian bytes for SQLite BLOB storage
• blob_to_vec — decode a SQLite BLOB back into a Vec<f32>

Provider Selection

Use create_provider to instantiate the appropriate provider based on the configuration:

let config = EmbeddingConfig::default(); // provider = "disabled"
let provider = create_provider(&config).unwrap();
assert_eq!(provider.model_name(), "disabled");

Retry Strategy

The OpenAI and Ollama providers use exponential backoff for transient errors:

• HTTP 429 (rate limited) and 5xx (server error) → retry
• HTTP 4xx (client error, not 429) → fail immediately
• Network errors → retry
• Backoff: 1s, 2s, 4s, 8s, 16s, 32s (capped at 2^5)

Structs

DisabledProvider

A no-op embedding provider that always returns errors.

LocalProvider

Embedding provider for local inference (fastembed on primary platforms, tract on musl/Intel Mac).

OllamaProvider

Embedding provider using a local Ollama instance.

OpenAIProvider

Embedding provider using the OpenAI API.

Traits

EmbeddingProvider

Trait for embedding providers.

Functions

blob_to_vec

Decode a BLOB back into a float vector.

config_to_fastembed_model 🔒

cosine_similarity

Compute cosine similarity between two embedding vectors.

create_provider

Create the appropriate EmbeddingProvider based on configuration.

embed_local_fastembed 🔒

embed_ollama 🔒

embed_openai 🔒

Call the OpenAI embeddings API with retry/backoff.

embed_query

Embed a single query text.

embed_texts

Embed a batch of texts using the configured provider.

parse_ollama_response 🔒

parse_openai_response 🔒

Parse the OpenAI embeddings API response JSON.

resolve_local_model 🔒

vec_to_blob

Encode a float vector as a BLOB (little-endian f32 bytes).