Module router 

Memory routing — classify recall queries and dispatch to the right backend.

The MemoryRouter trait is implemented by:

Type	Strategy
HeuristicRouter	Fast regex/keyword pattern matching. No LLM call.
LlmRouter	Uses an LLM classifier for high-accuracy routing.
HybridRouter	Runs HeuristicRouter first; escalates to LlmRouter when confidence is low.
AsyncMemoryRouter	Async wrapper over any MemoryRouter for use in async contexts.

Routing pipeline

1. HeuristicRouter classifies the query in < 1 ms using temporal-keyword detection and graph-relationship pattern matching.
2. If confidence >= threshold, the route is used directly.
3. Otherwise, HybridRouter forwards the query to LlmRouter for a second opinion.

Structs

HeuristicRouter

Heuristic-based memory router.

HybridRouter

Hybrid router: heuristic-first, escalates to LLM when confidence is low.

LlmRouter

LLM-based memory router.

RoutingDecision

Routing decision with confidence and optional LLM reasoning.

TemporalRange

Resolved datetime boundaries for a temporal query.

Enums

MemoryRoute

Classification of which memory backend(s) to query.

Traits

AsyncMemoryRouter

Async extension for LLM-capable routers.

MemoryRouter

Decides which memory backend(s) to query for a given input.

Functions

classify_graph_subgraph

Classify a query into the MAGMA edge types to use for subgraph-scoped BFS retrieval.

parse_route_str

Parse a route name string into a MemoryRoute, falling back to fallback on unknown values.

resolve_temporal_range

Resolve temporal keywords in query to a (after, before) datetime boundary pair.

strip_temporal_keywords

Strip matched temporal keywords from a query string before passing to FTS5.