Struct MemoryManager 

pub struct MemoryManager { /* private fields */ }

Orchestrates token counting and context compaction for conversation management.

MemoryManager is the main entry point for memory operations. It checks whether compaction is needed and delegates to the Compactor when the conversation exceeds configured token limits.

Implementations

impl MemoryManager

pub fn new( settings: &MemorySettings, cascade_config: Arc<AppConfig>, db_conn: Connection, ) -> Result<Self>

Create a new memory manager.

Arguments

• settings - Memory configuration (token limits, compaction ratio, etc.)
• cascade_config - llm-cascade AppConfig for summarization inference
• db_conn - SQLite connection for llm-cascade cooldown/logging

pub fn count_tokens(&self, state: &impl ConversationStateAccess) -> usize

Count the total tokens in a conversation state.

pub fn should_compact(&self, token_count: usize) -> bool

Check whether compaction should be triggered based on current token count.

Compaction is recommended when tokens exceed the configured limit.

pub async fn compact( &self, state: &mut impl ConversationStateAccess, ) -> Result<CompactionReport>

Attempt to compact the conversation if it exceeds the token limit.

If compaction is not needed, returns a no-op report. If compaction fails (e.g., summarization LLM error), returns the error without modifying the conversation state.

pub fn token_limit(&self) -> usize

Returns the configured context token limit.

pub fn token_counter(&self) -> &TokenCounter

Returns a reference to the underlying token counter.