Skip to main content

Module config

Module config 

Source

Structs§

Config
ConsolidationConfig
DatabaseConfig
DeduplicationConfig
EmbeddingConfig
HnswConfig
LlmConfig
SearchConfig
Hybrid-search blend parameters. rrf_k is the Reciprocal Rank Fusion smoothing constant (larger values flatten the rank-position weighting); vector_weight and sparse_weight scale the dense-vector and full-text contributions before fusion.
ServerConfig
TrainerConfig

Constants§

DEFAULT_EMBEDDING_HOST
DEFAULT_EMBEDDING_MODEL
DEFAULT_OLLAMA_HOST

Functions§

deterministic_provider_max_concurrent
disable_deterministic_provider_instrumentation
Disable instrumentation. Tests should call this in teardown to avoid leaking state across serially-run suites.
enable_deterministic_provider_instrumentation
Enable test-only instrumentation (counter + 20ms sleep) on the deterministic embedding provider. Off in production by default; tests enable it before reading concurrency counters.
expand_tilde
home_directory
reset_deterministic_provider_counters
resolve_project_dir
validate_dedup_threshold
Validates a configured deduplication threshold lies in the half-open cosine-similarity range (0.0, 1.0]. A threshold of 0.0 or below would treat unrelated memories as duplicates; above 1.0 is unreachable for normalized embeddings and signals a misconfiguration.
validate_fts_similarity_floor
Validates the consolidation FTS similarity floor. The floor is compared against a mean |bm25| rank (non-negative, unbounded above), so any finite non-negative value is meaningful. NaN/Inf are rejected explicitly — TOML 1.0 parses nan and inf, and a NaN floor would silently disable every comparison.
validate_search_config
Validates the hybrid-search blend parameters. rrf_k must be non-zero (it appears in the Reciprocal Rank Fusion denominator k + rank), both weights must be finite and non-negative, and at least one weight must be positive — two zero weights would cancel every contribution and yield an empty ranking regardless of how many results each index returns. NaN/Inf weights are rejected outright (NaN passes a naive < 0.0 range check).