Configuration Reference¶

All configuration is via environment variables. Defaults are tuned for “high‑alpha” search on real‑world manifests.

Canonical defaults are captured in docs/config_defaults.json and generated from src/config/ via scripts/update_config_reference.sh. When defaults change, regenerate that file and ensure this page stays in sync.

For end-to-end install/runtime composition patterns (binary + manifest + artifacts + models + SQL provider), see Modes & Combinations.

Core¶

DBT_MANIFEST_PATH – path to manifest.json (default: manifest.json)
DBT_NOVA_MANIFEST_URI – optional manifest URI (file://, http(s)://, dbfs://, s3://, gs://)
DBT_NOVA_MANIFEST_CACHE_DIR – optional local cache dir for remote manifests (default: <storage_root>/manifests)
DBT_NOVA_MANIFEST_REFRESH_SECS – refresh interval for remote manifests (0 = never refresh, default: 300)
DBT_NOVA_MANIFEST_MAX_BYTES – max bytes allowed for remote manifest fetches (0 = unlimited, default: 268435456)
DBT_NOVA_MANIFEST_HTTP_CONNECT_TIMEOUT_SECS – HTTP connect timeout for manifest fetches (0 = disabled, default: 10)
DBT_NOVA_MANIFEST_HTTP_TIMEOUT_SECS – HTTP request timeout for manifest fetches (0 = disabled, default: 120)
DBT_NOVA_MANIFEST_FETCH_TIMEOUT_SECS – total fetch deadline for manifest fetches (0 = disabled, default: 300)
DBT_NOVA_MANIFEST_ALLOW_HTTP – allow http:// manifest URIs (true|false, default: false)
DBT_NOVA_STORAGE_ARTIFACT_URI – optional URI to prebuilt storage archive (file://, s3://, gs://, dbfs://, http(s)://)
DBT_NOVA_METADATA_ARTIFACT_URI – optional URI to prebuilt metadata contract JSON (required with DBT_NOVA_STORAGE_ARTIFACT_URI)
DBT_NOVA_MODELS_ARTIFACT_URI – optional URI to prebuilt models archive
DBT_NOVA_BOOTSTRAP_URI – optional URI to a bootstrap contract JSON that can populate manifest_uri, storage_instance_id, and prebuilt artifact URIs (same supported schemes as prebuilt artifact URIs)
DBT_NOVA_ARTIFACTS_CACHE_DIR – optional cache dir for downloaded artifact archives (default: <storage_root>/artifacts)
DBT_NOVA_ARTIFACT_FETCH_POLICY – artifact fetch policy (if_missing, always, never; default: if_missing)
DBT_NOVA_ARTIFACT_TIMEOUT_SECS – fetch timeout for remote artifact downloads (0 = disabled, default: 300)
DBT_NOVA_ARTIFACT_ALLOW_HTTP – allow http:// artifact URIs (true|false, default: false)
DBT_NOVA_STRICT_SCHEMA – fail build if schema files are missing or invalid (true|false, default: false; forced true in CI)
DBT_NOVA_S3_MODE – S3 fetch mode (https or sdk, default: https)
DBT_NOVA_GCS_MODE – GCS fetch mode (https or sdk, default: https)
DBT_NOVA_RECIPES_DIR – manifest original_file_path prefix used to discover recipe analysis nodes (default: analyses/recipes). Recipe SQL is resolved from manifest compiled_code (or raw_code fallback). Recipes are documented in Analysis Recipes.
DBT_NOVA_LOG / RUST_LOG – enable structured logs to stderr (e.g., info, debug, trace)
DBT_NOVA_DISABLE_TOOL_SCHEMAS – strip JSON schema hints from MCP tools (useful for strict clients like Gemini; see MCP Clients)
DBT_NOVA_SQL_PROVIDER – SQL backend for execute_sql (databricks, bigquery, or duckdb, default: databricks)
DBT_NOVA_GCP_PROJECT_ID – shared Google project id alias (used by BigQuery fallback resolution)
DBT_NOVA_GCP_ACCESS_TOKEN – shared Google OAuth access token alias (used by BigQuery fallback resolution)
DBT_NOVA_BIGQUERY_PROJECT_ID – BigQuery project id when DBT_NOVA_SQL_PROVIDER=bigquery (falls back to DBT_NOVA_GCP_PROJECT_ID, GOOGLE_CLOUD_PROJECT, GCP_PROJECT_ID)
DBT_NOVA_BIGQUERY_ACCESS_TOKEN – OAuth access token for BigQuery when DBT_NOVA_SQL_PROVIDER=bigquery (falls back to DBT_NOVA_GCP_ACCESS_TOKEN, GCP_ACCESS_TOKEN, GOOGLE_OAUTH_ACCESS_TOKEN, GOOGLE_APPLICATION_CREDENTIALS, or gcloud ADC)
DBT_NOVA_BIGQUERY_LOCATION – optional BigQuery location for execute_sql and provider preflight
DBT_NOVA_BIGQUERY_TIMEOUT_MS – HTTP timeout for BigQuery API requests (default: 30000)
DBT_NOVA_BIGQUERY_TOKEN_CACHE_TTL_SECS – cache TTL for BigQuery auth token + HTTP client reuse (default: 3000, minimum: 60)
DBT_NOVA_DUCKDB_PATH – required DuckDB database file when DBT_NOVA_SQL_PROVIDER=duckdb
DBT_NOVA_DUCKDB_FILE_SEARCH_PATH – optional DuckDB file_search_path used for external file-backed objects when DBT_NOVA_SQL_PROVIDER=duckdb
DBT_NOVA_DUCKDB_POOL_MAX_SIZE – optional max pooled DuckDB connections per (duckdb_path,file_search_path) key (default: falls back to DBT_NOVA_SQL_MAX_CONCURRENT, then 10)
DATABRICKS_HOST – Databricks workspace URL for dbfs:// manifests and execute_sql
DATABRICKS_ACCESS_TOKEN – Databricks access token for dbfs:// and execute_sql

For auth details by source, see docs/configuration/manifest-sources.md.

Remote manifest notes: - s3:// and gs:// are fetched over HTTPS by default (public or presigned URLs). - Optional overrides: DBT_NOVA_S3_ENDPOINT, DBT_NOVA_GCS_ENDPOINT. - dbfs:// requires DATABRICKS_HOST + DATABRICKS_ACCESS_TOKEN. - To use SDK credentials, set DBT_NOVA_S3_MODE=sdk or DBT_NOVA_GCS_MODE=sdk (SDKs are included by default builds). - To force HTTPS for public/presigned URLs, set DBT_NOVA_S3_MODE=https or DBT_NOVA_GCS_MODE=https. - To allow insecure http:// manifests (not recommended), set DBT_NOVA_MANIFEST_ALLOW_HTTP=true. - To allow insecure http:// prebuilt artifact URIs (not recommended), set DBT_NOVA_ARTIFACT_ALLOW_HTTP=true. - Bootstrap precedence is deterministic: explicit env vars override bootstrap values, and bootstrap values override defaults.

Storage¶

Data Loss Risk

Setting DBT_NOVA_CLEANUP_STORAGE_ON_START=true deletes all cached indexes on startup. Use only for development or when you need a clean slate.

DBT_NOVA_STORAGE_DIR – base directory for on‑disk storage (default: .dbt-nova)
DBT_NOVA_STORAGE_INSTANCE_ID – optional instance id (auto‑generated when unset)
DBT_NOVA_CLEANUP_STORAGE_ON_START – delete instance dir on startup (true|false, default: false)
DBT_NOVA_STORAGE_MAX_INSTANCES – max instance dirs to retain (0 = unlimited, default: 3)
DBT_NOVA_STORAGE_MIN_VERSIONS – minimum manifest versions retained per instance (default: 2)
DBT_NOVA_STORAGE_MAX_BYTES – max total bytes across instances (0 = unlimited, default: 5368709120)
DBT_NOVA_STORAGE_BUILD_LOCK_WAIT_SECS – max seconds to wait for another process to finish building (default: 300)
DBT_NOVA_STORAGE_READ_ONLY – do not build indexes (true|false, default: false)
DBT_NOVA_ENTITY_CACHE_SIZE – max entities cached in memory (0 disables, default: 1000)
DBT_NOVA_EMBEDDINGS_CACHE_DIR – embeddings cache directory (default: models/ next to executable if present, else <storage_root>/.fastembed_cache)

Embeddings cache resolution order when DBT_NOVA_EMBEDDINGS_CACHE_DIR is unset:

models/ next to the active executable
~/.local/bin/models (if present)
<storage_root>/.fastembed_cache

Notes: - The entity cache backend is currently fixed to moka (no env override).

Instance directories live under <storage_root>/instances/<instance_id>.

Safety Limits¶

DBT_NOVA_BATCH_GET_MAX_ITEMS – max ids accepted by batch_get_entities (0 = unlimited, default: 5000)
DBT_NOVA_MAX_PAGE_SIZE – max results per page (default: 2000)
DBT_NOVA_MAX_OFFSET – max pagination offset (default: 10000)
DBT_NOVA_MAX_QUERY_LENGTH – max search query length (default: 2000)
DBT_NOVA_MAX_PATH_PATTERN_LENGTH – max path pattern length (default: 1000)
DBT_NOVA_MAX_SQL_CHUNK_BYTES – max bytes of SQL indexed per field (default: 262144)
DBT_NOVA_EMBEDDINGS_MAX_DECOMPRESSED_BYTES – max decompressed bytes for embedding caches (0 = unlimited, default: 4294967296)
DBT_NOVA_TOOL_RATE_LIMITS – per-tool rate limits (default: search=60,execute_sql=20,default=120)
DBT_NOVA_TOOL_RATE_LIMIT_WINDOW_SECS – rate limit window seconds (default: 60)
DBT_NOVA_SQL_MAX_ROW_LIMIT – max rows accepted by execute_sql (0 = unlimited, default: 10000)
DBT_NOVA_SQL_MAX_BYTE_LIMIT – max bytes accepted by execute_sql (0 = unlimited, default: 100000000)
DBT_NOVA_SQL_MAX_CHUNKS – max result chunks accepted by execute_sql (0 = unlimited, default: 100)
DBT_NOVA_SQL_MAX_POLL_SECONDS – max polling duration accepted by execute_sql (0 = unlimited, default: 900)
DBT_NOVA_SQL_MIN_POLL_INTERVAL_MS – minimum poll interval accepted by execute_sql (0 disables floor, default: 200)
DBT_NOVA_SQL_MAX_CONCURRENT – max concurrent SQL executions across execute_sql and run_recipe (0 = unlimited, default: 10)
DBT_NOVA_SQL_MAX_QUEUE – max queued SQL executions while all slots are busy (default: 20)
DBT_NOVA_SQL_QUEUE_TIMEOUT_MS – max wait time for a SQL execution slot (0 disables timeout, default: 30000)

Lineage¶

DBT_NOVA_DEFAULT_CONFIDENCE – column lineage confidence (high|medium|low, default: medium)
DBT_NOVA_LEVENSHTEIN_THRESHOLD – similarity threshold (default: 0.75)
DBT_NOVA_MIN_PREFIX_SUFFIX_LENGTH – min length for prefix/suffix matches (default: 2)
DBT_NOVA_MIN_LEVENSHTEIN_LENGTH – min length to run Levenshtein (default: 3)
DBT_NOVA_SQL_PROXIMITY_MAX_DISTANCE – max SQL proximity distance (default: 100)
DBT_NOVA_MAX_LINEAGE_RESULTS – max column lineage results (default: 10000)
DBT_NOVA_COLUMN_LINEAGE_MAX_DEPTH – max column lineage depth (default: 100)
DBT_NOVA_COLUMN_LINEAGE_MAX_CANDIDATES – max column match candidates before capping (default: 10000)
DBT_NOVA_COLUMN_LINEAGE_PRECOMPUTE – precompute SQL aliases for column lineage (true|false, default: true)
DBT_NOVA_MAX_ENTITY_LINEAGE_RESULTS – max entity lineage results (default: 10000)
DBT_NOVA_MAX_LINEAGE_DEPTH – max entity lineage depth (default: 200)
DBT_NOVA_LINEAGE_CACHE_SIZE – cache size for entity lineage responses (0 disables, default: 2048)

Note: DBT_NOVA_MAX_LINEAGE_RESULTS applies to column lineage (get_column_lineage), while DBT_NOVA_MAX_ENTITY_LINEAGE_RESULTS applies to entity lineage (get_lineage).

Tantivy / Lexical Search¶

DBT_NOVA_DEFAULT_LIMIT – fallback result limit when limit=0 (default: 50)
DBT_NOVA_MIN_WORD_LENGTH – minimum word length for indexing (default: 2)
DBT_NOVA_INDEX_DIR – Tantivy index directory name inside storage (default: index)
DBT_NOVA_INDEX_WRITER_HEAP_BYTES – Tantivy writer heap size (default: 128000000)
DBT_NOVA_SEARCH_DEDUP_FETCH_MULTIPLIER – fetch multiplier for de‑duping (default: 8)
DBT_NOVA_SEARCH_ENABLE_NGRAM – enable n‑gram indexing (default: true)
DBT_NOVA_SEARCH_NGRAM_MIN – min n‑gram size (default: 3)
DBT_NOVA_SEARCH_NGRAM_MAX – max n‑gram size (default: 3)
DBT_NOVA_SEARCH_NGRAM_BOOST – n‑gram boost (default: 0.35)
DBT_NOVA_FUZZY_MIN_LENGTH – min term length for fuzzy (default: 4)
DBT_NOVA_FUZZY_MID_LENGTH – medium fuzzy threshold (default: 7)
DBT_NOVA_FUZZY_MAX_DISTANCE – max edit distance (default: 2)
DBT_NOVA_SEARCH_HIGHLIGHT_MAX_CHARS – max snippet length (default: 240)
DBT_NOVA_SEARCH_HIGHLIGHT_MAX_FIELDS – max highlighted fields per result (default: 5)
DBT_NOVA_SEARCH_HIGHLIGHT_FORMAT – text or html (default: text)
DBT_NOVA_SEARCH_ENABLE_SUGGESTIONS – enable suggestions (default: true)
DBT_NOVA_SEARCH_SUGGESTIONS_LIMIT – max suggestions (default: 7)
DBT_NOVA_SEARCH_ENABLE_RRF – enable RRF fusion (default: true)
DBT_NOVA_SEARCH_RRF_K – RRF smoothing constant (default: 60)
DBT_NOVA_SEARCH_RRF_OVERFETCH – overfetch multiplier (default: 3)
DBT_NOVA_SEARCH_TIMEOUT_MS – search timeout in ms (0 disables, default: 30000)
DBT_NOVA_SEARCH_MAX_CONCURRENT – max concurrent search requests (0 = unlimited, default: 4)
DBT_NOVA_SEARCH_MAX_QUEUE – max queued searches when saturated (default: 8)

Embeddings (Dense + Sparse)¶

High Memory Usage

Enabling dense vectors (DBT_NOVA_SEARCH_ENABLE_VECTOR=true) requires ~2 GB RAM for embeddings. Disable on memory-constrained systems.

DBT_NOVA_SEARCH_ENABLE_VECTOR – enable dense vectors (default: true)
DBT_NOVA_SEARCH_VECTOR_TOP_K – max vector hits before fusion (default: 200)
DBT_NOVA_SEARCH_VECTOR_MAX_CHARS – max chars in embedding text (default: 4000)
DBT_NOVA_SEARCH_ENABLE_VECTOR_ANN – enable ANN buckets (default: true)
DBT_NOVA_SEARCH_ENABLE_VECTOR_QUANTIZATION – enable 8‑bit quantization (default: false)
DBT_NOVA_SEARCH_VECTOR_ANN_BITS – ANN hash bits (default: 16)
DBT_NOVA_SEARCH_VECTOR_ANN_HAMMING – ANN Hamming radius (default: 1)
DBT_NOVA_SEARCH_VECTOR_ANN_MAX_CANDIDATES – max ANN candidates (default: 5000)
DBT_NOVA_SEARCH_VECTOR_ANN_MIN_CANDIDATES – min before full scan (default: 200)
DBT_NOVA_SEARCH_EMBEDDING_BATCH_SIZE – embedding batch size (default: 128)
DBT_NOVA_EMBEDDING_MODEL – embedding model name (default: intfloat/multilingual-e5-base)
DBT_NOVA_SEARCH_ENABLE_SPARSE – enable sparse vectors (default: true)
DBT_NOVA_SEARCH_SPARSE_TOP_K – max sparse hits before fusion (default: 200)
DBT_NOVA_SEARCH_ENABLE_RERANKER – enable cross‑encoder reranker (default: true)
DBT_NOVA_RERANKER_MODEL – reranker model (default: jinaai/jina-reranker-v2-base-multilingual)
DBT_NOVA_SEARCH_RERANK_TOP_N – max results reranked (default: 20)

Proxy validation: - HTTP_PROXY, HTTPS_PROXY, ALL_PROXY (and lowercase variants) must be absolute URLs when set. - Invalid proxy values fail fast during embeddings/reranker initialization.

Persona Defaults + Overrides¶

DBT_NOVA_SEARCH_DEFAULT_PERSONA – default persona if request omits persona
DBT_NOVA_SEARCH_PERSONA_ANALYST_WEIGHTS – comma‑separated overrides (e.g. vector=1.6,docs=1.3)
DBT_NOVA_SEARCH_PERSONA_ENGINEER_WEIGHTS – overrides for engineer persona
DBT_NOVA_SEARCH_PERSONA_GOVERNANCE_WEIGHTS – overrides for governance persona
DBT_NOVA_SEARCH_PERSONA_DEFAULT_WEIGHTS – overrides for default persona

Aliases: docs and documentation are both accepted in persona weight overrides.

Analyst semantic multipliers (advanced tuning): - DBT_NOVA_SEARCH_ANALYST_METRIC_DEF_MULTIPLIER (default: 1.09) - DBT_NOVA_SEARCH_ANALYST_MEASURE_DEF_MULTIPLIER (default: 1.05) - DBT_NOVA_SEARCH_ANALYST_GRAIN_MULTIPLIER (default: 1.05) - DBT_NOVA_SEARCH_ANALYST_TIME_FIELD_MULTIPLIER (default: 1.03) - DBT_NOVA_SEARCH_ANALYST_DIM_OVERLAP_ONE_MULTIPLIER (default: 1.03) - DBT_NOVA_SEARCH_ANALYST_DIM_OVERLAP_TWO_MULTIPLIER (default: 1.06) - DBT_NOVA_SEARCH_ANALYST_DIM_OVERLAP_THREE_PLUS_MULTIPLIER (default: 1.09) - DBT_NOVA_SEARCH_ANALYST_MISSING_METRIC_MEASURE_MULTIPLIER (default: 0.96) - DBT_NOVA_SEARCH_ANALYST_MISSING_GRAIN_MULTIPLIER (default: 0.97) - DBT_NOVA_SEARCH_ANALYST_MIN_MULTIPLIER (default: 0.85) - DBT_NOVA_SEARCH_ANALYST_MAX_MULTIPLIER (default: 1.35)

Metadata Scoring¶

Metadata scoring uses the weight defaults under metadata_score.persona_weights in docs/config_defaults.json (sourced from src/config/metadata_score.rs). There are currently no environment overrides for these weights.

Governance Gate Policy¶

DBT_NOVA_GOV_GATE_PROFILE – preset governance gate policy (strict|standard|advisory)
DBT_NOVA_GOV_GATE_POLICY – full policy override as JSON
DBT_NOVA_GOV_GATE_MIN_METADATA_SCORE – minimum metadata score threshold (0..=100)
DBT_NOVA_GOV_GATE_MIN_DOC_COVERAGE_PCT – minimum documentation coverage threshold (0..=100)
DBT_NOVA_GOV_GATE_REQUIRE_TESTS – require tests to pass governance gate (true|false)
DBT_NOVA_GOV_GATE_REQUIRE_OWNER – require owner metadata to pass governance gate (true|false)
DBT_NOVA_GOV_GATE_REQUIRE_REQUIRED_FIELDS – enforce DBT_NOVA_GOV_REQUIRED_FIELDS checks (true|false)
DBT_NOVA_GOV_GATE_REQUIRE_COMPLIANCE_FOR_PII – require compliance tags when PII is declared (true|false)
DBT_NOVA_GOV_GATE_BLOCK_ON_FAILURE – return blocking fail status on gate failure (true), or advisory-only status (false)

Circuit Breakers¶

DBT_NOVA_SEARCH_CIRCUIT_FAILURE_THRESHOLD – failures before opening (default: 3)
DBT_NOVA_SEARCH_CIRCUIT_OPEN_SECONDS – open duration seconds (default: 60)

Search Boosts & Nova Ranking¶

These variables control lexical boosts and Nova‑specific ranking multipliers. Full rationale and defaults are documented in docs/features/search-ranking.md.

Base field boosts: - DBT_NOVA_ALIAS_BOOST (default: 18.0) - DBT_NOVA_NAME_BOOST (default: 12.0) - DBT_NOVA_DESCRIPTION_BOOST (default: 6.0) - DBT_NOVA_COLUMN_BOOST (default: 4.0) - DBT_NOVA_TAG_BOOST (default: 3.0) - DBT_NOVA_PATH_BOOST (default: 2.0) - DBT_NOVA_CODE_BOOST (default: 1.5)

Nova meta boosts: - DBT_NOVA_META_SYNONYMS_BOOST (default: 7.0) - DBT_NOVA_META_DOMAINS_BOOST (default: 4.0) - DBT_NOVA_META_USE_CASES_BOOST (default: 4.0) - DBT_NOVA_META_MEASURES_BOOST (default: 8.0) - DBT_NOVA_META_METRIC_BOOST (default: 10.0) - DBT_NOVA_META_SENSITIVITY_BOOST (default: 6.0) - DBT_NOVA_META_PII_BOOST (default: 8.0) - DBT_NOVA_META_COMPLIANCE_BOOST (default: 6.0)

Post‑retrieval tuning: - DBT_NOVA_SEARCH_STAGING_DEBOOST_FACTOR (default: 0.6) - DBT_NOVA_SEARCH_MEASURE_MATCH_MULTIPLIER (default: 1.15) - DBT_NOVA_SEARCH_METRIC_MATCH_MULTIPLIER (default: 1.2) - DBT_NOVA_SEARCH_SYNONYM_MATCH_MULTIPLIER (default: 1.2) - DBT_NOVA_SEARCH_CANONICAL_MATCH_MULTIPLIER (default: 1.08) - DBT_NOVA_SEARCH_CANONICAL_META_MATCH_MULTIPLIER (default: 1.35) - DBT_NOVA_SEARCH_CANONICAL_META_MATCH_BONUS (default: 2.5) - DBT_NOVA_SEARCH_ENGINEER_EXACT_MATCH_MULTIPLIER (default: 2.0)

Staging deboost behavior: - DBT_NOVA_SEARCH_STAGING_DEBOOST_FACTOR applies when an entity matches a configured layer rule with layer name staging, stage, or stg (case-insensitive). - Configure layer mapping with DBT_NOVA_LAYER_RULES_JSON.

SQL Providers¶

execute_sql uses the SQL provider configured by DBT_NOVA_SQL_PROVIDER. Supported providers:

databricks (default): requires DATABRICKS_HOST, DATABRICKS_ACCESS_TOKEN, and a warehouse id (DATABRICKS_HTTP_PATH or DATABRICKS_SQL_WAREHOUSE_ID).
bigquery: requires a project id (DBT_NOVA_BIGQUERY_PROJECT_ID, DBT_NOVA_GCP_PROJECT_ID, or GOOGLE_CLOUD_PROJECT) and credentials from one of:
OAuth token env: DBT_NOVA_BIGQUERY_ACCESS_TOKEN, DBT_NOVA_GCP_ACCESS_TOKEN, GCP_ACCESS_TOKEN, GOOGLE_OAUTH_ACCESS_TOKEN
Service-account key path: GOOGLE_APPLICATION_CREDENTIALS
gcloud ADC (gcloud auth application-default login)
duckdb: requires DBT_NOVA_DUCKDB_PATH and executes queries against that file in read-only mode. Optional DBT_NOVA_DUCKDB_FILE_SEARCH_PATH configures DuckDB file_search_path for external file-backed objects.

Databricks runtime tuning env vars: - DATABRICKS_WAIT_TIMEOUT_S (default: 10) - DATABRICKS_POLL_INTERVAL_MS (default: 1000) - DATABRICKS_MAX_POLL_SECONDS (default: 600) - DATABRICKS_TIMEOUT_MS (derived from wait timeout + 5 seconds, min 30000) - DATABRICKS_MAX_GET_RETRIES (default: 2)

Provider diagnostics are available through execute_sql with preflight_only=true plus optional preflight_catalog, preflight_schema, and preflight_relation. Object-level preflight checks (preflight_catalog, preflight_schema, preflight_relation) are treated as available only when the probe query returns at least one row.

DuckDB notes: - Named parameters are supported and rewritten to positional binds. - parameter_types is not supported for DuckDB v1 (pass scalar values via parameters only). - Connections are pooled per process and per (duckdb_path,file_search_path) key.

When provided by callers, row_limit, byte_limit, max_chunks, and max_poll_seconds are clamped to the configured DBT_NOVA_SQL_MAX_* values. poll_interval_ms is raised to DBT_NOVA_SQL_MIN_POLL_INTERVAL_MS when below the configured minimum.

See MCP Client Configs for required variables.

Search boost and Nova ranking tuning variables are documented in Search Ranking.