Changelog

All notable changes to this project will be documented here.

Version 0.0.2

Support for arbitrary sentence-transformers models using SentenceTransformerEngine
New LLMEngine for pointwise reranking and query decomposition via vLLM or any OpenAI-compatible endpoint, with optional runtime-supplied prompts
gRPC server (opt-in via --grpc) alongside the REST endpoints, sharing the same processor registry; new routir.client subpackage with a unified AsyncClient/Client that auto-discovers gRPC from a REST endpoint and falls back to REST when gRPC is unreachable
Pipeline-level result cache: enable via top-level pipeline_cache / pipeline_cache_ttl / pipeline_cache_redis_url in the config; keyed on the canonical pipeline AST + query + collection + the subset of runtime_kwargs referenced by the pipeline, so whitespace and alias-vs-expansion collapse to one entry
server_imports now also proxies remote content services (collections); a new local cache (relay_content_cache) avoids re-fetching document text on every rerank
Pre-defined pipeline_aliases in the config: name common pipeline DSL fragments and reuse them at request time
--api_key (or ROUTIR_API_KEY env var) enables Bearer-token authentication on all REST and gRPC endpoints except /ping; CORS preflight requests are exempt
collection field is optional on /pipeline; required only when the pipeline contains a reranking stage
ASCII banner with package version printed on startup