Expand description
Search layer facade.
This module provides the search infrastructure for cass, including:
query: Query parsing, execution, and caching for Tantivy-based full-text search.tantivy: Tantivy index creation, schema management, and document indexing.embedder: Embedder trait for semantic search (hash and ML implementations).embedder_registry: Embedder registry for model selection (bd-2mbe).hash_embedder: FNV-1a feature hashing embedder (deterministic fallback).fastembed_embedder: FastEmbed-backed ML embedder (MiniLM).reranker: Reranker trait for cross-encoder reranking of search results.reranker_registry: Reranker registry for model selection with bake-off support.fastembed_reranker: FastEmbed-backed cross-encoder reranker (ms-marco-MiniLM-L-6-v2).daemon_client: Daemon client wrappers for warm embedder/reranker (bd-1lps).model_manager: Semantic model detection + context wiring (no downloads).model_download: Model download system with consent, verification, and atomic install.policy: Semantic policy contract: model defaults, tiers, budgets, invalidation.semantic_manifest: Durable semantic asset manifests, backlog ledger, and checkpoints.canonicalize: Text preprocessing for consistent embedding input.ann_index: HNSW-based approximate nearest neighbor index (Opt 9).two_tier_search: Two-tier progressive search with fast/quality embeddings (bd-3dcw).pack_planner: Deterministic answer-pack evidence selection core.
Modulesยง
- ann_
index - Approximate nearest-neighbor (ANN) reporting types.
- asset_
state - Shared search asset state evaluation for status, health, and fail-open search planning.
- canonicalize
- Text canonicalization for consistent embedding input.
- daemon_
client - Daemon client integration re-exports.
- embedder
- Embedder trait and types for semantic search.
- embedder_
registry - Embedder registry for model selection (bd-2mbe).
- fastembed_
embedder - FastEmbed-based ML embedders.
- fastembed_
reranker - FastEmbed-based cross-encoder reranker (ms-marco-MiniLM-L-6-v2).
- hash_
embedder - FNV-1a feature hashing embedder.
- model_
download - Model download and management system.
- model_
manager - Semantic model management (local-only detection).
- pack_
planner - Deterministic answer-pack evidence selection.
- policy
- Semantic policy contract for cass hybrid search.
- query
- reranker
- Reranker trait and types for cross-encoder reranking.
- reranker_
registry - Reranker registry for model selection.
- runtime_
optimizations - Runtime-toggleable performance optimizations.
- semantic_
manifest - Durable semantic asset manifest, backlog ledger, and resumable checkpoints.
- tantivy
- two_
tier_ search - Two-tier progressive search for session search (bd-3dcw, bd-2fu7e).
- vector_
index - Vector index facade for cass.