Module memory 

Memory management module for conversation context and user memory.

This module provides utilities for:

• Building agent context with memory
• Formatting memory for LLM prompts
• Managing conversation history windows

User memory facts and preferences are stored in the database (PostgresClient). This module provides utilities for working with that stored memory.

Structs

MemoryConfig

Configuration for the in-memory session store.

MemorySession

A tenant-scoped conversation session stored in memory.

MemoryStore

In-memory LRU session store with per-tenant isolation and TTL.

Enums

MemoryError

Errors from session store operations.

Constants

DEFAULT_HISTORY_WINDOW

Default number of recent messages to include in context.

MAX_FACTS_IN_PROMPT

Maximum number of facts to include in a prompt to avoid token overflow.

MAX_PREFERENCES_IN_PROMPT

Maximum number of preferences to include in a prompt.

Functions

build_context

Builds an agent context from components.

cleanup_expired

Remove expired sessions; returns the number removed.

estimate_tokens

Estimates token count for a message.

filter_facts_by_category

Filters memory facts by category.

filter_preferences_by_category

Filters preferences by category.

format_memory_for_prompt

Formats user memory into a string suitable for inclusion in system prompts.

format_preferences_compact

Formats user preferences into a compact string for prompt inclusion.

lru_evict

Evict least-recently-used entries until order.len() <= max_sessions.

lru_get

Promote key to most-recently-used (end of order); returns whether it was present.

lru_put

Insert or promote key to most-recently-used.

session_key

Composite map key for tenant-scoped sessions.

truncate_history

Truncates conversation history to a window of recent messages.

truncate_history_to_tokens

Truncates history to fit within a token budget.

ttl_expired

Returns true when now - created_at > ttl_secs (TTL disabled when ttl_secs == 0).