Semantic Router

Intent-based routing using embedding similarity.

`SemanticRouter`

Route incoming queries to named handlers by embedding similarity.

Usage:

python

router = SemanticRouter(
    routes=[
        Route("weather", "Weather questions", examples=["Will it rain?"]),
        Route("travel", "Travel planning", examples=["Plan a trip to Rome"]),
    ],
    embed_fn=llm_service.embed,
    min_confidence=0.6,
    fallback_route="travel",
)
await router.compile()   # precompute embeddings (called at startup)

match = await router.route("Is it going to snow?")
print(match.route_name)   # "weather"

`Route`

A named route with description and optional example utterances.

Usage:

python

Route(
    name="weather",
    description="Questions about weather, forecasts, temperature",
    examples=["Will it rain?", "What is the temperature in Paris?"],
)

Parameters:

Name	Type	Description
`name`	`str`	Unique route identifier.
`description`	`str`	Human-readable description of what this route handles.
`examples`	`list[str]`	Optional list of example utterances used to compute the route centroid embedding.

`RouteMatch`

Result of a SemanticRouter.route() call.

Parameters:

Name	Type	Description
`route_name`	`str`	Name of the matched route (also accessible as `route` for API compatibility).
`confidence`	`float`	Cosine-similarity score of the best match.
`method`	`Literal['embedding', 'llm_fallback', 'default']`	How the match was determined: `"embedding"`, `"llm_fallback"`, or `"default"`.