Struct SemanticService 

pub struct SemanticService {
    pub provider: Box<dyn EmbeddingProvider>,
    /* private fields */
}

Main semantic embedding service for the memory system

Coordinates embedding generation, storage, and semantic search across episodes and patterns. Integrates with the existing storage backends.

Fields

provider: Box<dyn EmbeddingProvider>

Embedding provider for text-to-vector conversion

Implementations

impl SemanticService

pub fn new( provider: Box<dyn EmbeddingProvider>, storage: Box<dyn EmbeddingStorageBackend>, config: EmbeddingConfig, ) -> Self

Create a new semantic service with the specified provider and storage

pub fn config(&self) -> &EmbeddingConfig

Get the embedding configuration

pub async fn with_local_provider( storage: Box<dyn EmbeddingStorageBackend>, config: EmbeddingConfig, ) -> Result<Self>

Get the embedding provider type

pub async fn default(storage: Box<dyn EmbeddingStorageBackend>) -> Result<Self>

Create a semantic service with default local provider

pub async fn with_fallback( storage: Box<dyn EmbeddingStorageBackend>, config: EmbeddingConfig, ) -> Result<Self>

Create a semantic service with automatic provider fallback

Tries providers in order: Local → OpenAI → Mock (with warnings) This ensures maximum reliability by falling back to simpler options if preferred ones fail.

pub async fn embed_episode(&self, episode: &Episode) -> Result<Vec<f32>>

Generate and store embedding for an episode

Creates a semantic representation of the episode by combining:

• Task description
• Context information (domain, language, framework)
• Key execution steps
• Outcome summary

pub async fn embed_pattern(&self, pattern: &Pattern) -> Result<Vec<f32>>

Generate and store embedding for a pattern

Creates a semantic representation based on:

• Pattern description
• Context where the pattern was extracted
• Pattern metadata and effectiveness metrics

pub async fn find_similar_episodes( &self, query: &str, context: &TaskContext, limit: usize, ) -> Result<Vec<SimilaritySearchResult<Episode>>>

Find semantically similar episodes for a query

Uses vector similarity to find episodes that are semantically related to the query, going beyond keyword matching to understand meaning.

pub async fn find_similar_patterns( &self, context: &TaskContext, limit: usize, ) -> Result<Vec<SimilaritySearchResult<Pattern>>>

Find semantically similar patterns for a context

Identifies patterns that are semantically relevant to the given context, enabling better pattern reuse and recommendation.

pub async fn text_similarity(&self, text1: &str, text2: &str) -> Result<f32>

Calculate similarity between two texts

pub async fn find_episodes_by_embedding( &self, embedding: Vec<f32>, limit: usize, threshold: f32, ) -> Result<Vec<SimilaritySearchResult<Episode>>>

Find episodes similar to a pre-computed embedding vector

This method allows searching with a pre-computed embedding, useful when the embedding has been generated externally or cached.

Arguments

• embedding - Pre-computed embedding vector to search with
• limit - Maximum number of results to return
• threshold - Minimum similarity score (0.0-1.0)

Returns

Vector of similar episodes with their similarity scores

pub async fn find_patterns_by_embedding( &self, embedding: Vec<f32>, limit: usize, threshold: f32, ) -> Result<Vec<SimilaritySearchResult<Pattern>>>

Find patterns similar to a pre-computed embedding vector

This method allows searching with a pre-computed embedding, useful when the embedding has been generated externally or cached.

Arguments

• embedding - Pre-computed embedding vector to search with
• limit - Maximum number of results to return
• threshold - Minimum similarity score (0.0-1.0)

Returns

Vector of similar patterns with their similarity scores

pub async fn get_embeddings_batch( &self, episode_ids: &[Uuid], ) -> Result<Vec<Option<Vec<f32>>>>

Get embeddings for multiple episodes in batch

This method retrieves embeddings for multiple episode IDs efficiently. For backends that don’t support batch operations, it falls back to individual lookups.