Struct MemoryEngine 

pub struct MemoryEngine { /* private fields */ }

The core memory engine. Orchestrates the full add/search/get/update/delete/history pipeline.

Implementations

impl MemoryEngine

pub async fn add( &self, messages: &[ChatMessage], user_id: Option<&str>, agent_id: Option<&str>, run_id: Option<&str>, metadata: Option<&Value>, infer: bool, ) -> Result<AddResult>

Add memories from a conversation.

When infer is true (the default), the LLM extracts facts from the conversation, deduplicates them against existing memories, and decides whether to add, update, or delete.

When infer is false, each message’s content is stored directly as a new memory without any LLM processing — useful for importing raw text.

metadata is an optional JSON object stored under payload.metadata.

pub async fn add_with_options( &self, messages: &[ChatMessage], opts: &AddOptions<'_>, ) -> Result<AddResult>

Add memories using structured options.

impl MemoryEngine

pub async fn get(&self, memory_id: Uuid) -> Result<MemoryItem>

pub async fn get_all( &self, user_id: Option<&str>, agent_id: Option<&str>, run_id: Option<&str>, filters: Option<&Value>, limit: Option<usize>, ) -> Result<Vec<MemoryItem>>

pub async fn update(&self, memory_id: Uuid, new_text: &str) -> Result<()>

pub async fn delete(&self, memory_id: Uuid) -> Result<()>

pub async fn delete_all( &self, user_id: Option<&str>, agent_id: Option<&str>, run_id: Option<&str>, ) -> Result<()>

pub async fn history(&self, memory_id: Uuid) -> Result<Vec<MemoryEvent>>

pub async fn reset(&self) -> Result<()>

impl MemoryEngine

pub async fn new(config: MemoryEngineConfig) -> Result<Self>

impl MemoryEngine

pub async fn search( &self, query: &str, user_id: Option<&str>, agent_id: Option<&str>, run_id: Option<&str>, limit: usize, filters: Option<&Value>, rerank: bool, threshold: Option<f32>, task_type: Option<&str>, ) -> Result<SearchResult>

Search memories by semantic similarity.

filters is an optional JSON object evaluated against payload.metadata using the filter DSL (simple equality, operators, AND/OR/NOT).

When a reranker is configured and rerank is true, the engine over-fetches candidates by top_k_multiplier and then reranks them down to limit.

pub async fn search_with_options( &self, query: &str, opts: &SearchOptions<'_>, ) -> Result<SearchResult>

Search memories using structured options.