Jorvis Environment Variables Reference

Feature Flags

Variable	Type	Default	Description
`JORVIS_PARALLEL_QUERIES`	boolean	`false`	Enable parallel query execution
`JORVIS_FEATURE_WEB_SEARCH`	boolean	`false`	Enable web search fallback
`JORVIS_AUDIT_STORAGE_ENABLED`	boolean	`false`	Enable audit storage (ADR-0007)
`JORVIS_RRF_K`	number	`60`	RRF fusion constant for hybrid retrieval
`JORVIS_HYBRID_RAG_ENABLED`	boolean	`false`	Enable hybrid RAG (vector + keyword)
`JORVIS_FK_EXPANSION_ENABLED`	boolean	`false`	Enable FK-based table expansion
`JORVIS_RERANKER_ENABLED`	boolean	`false`	Enable deterministic reranker
`JORVIS_SEMANTIC_CACHE_ENABLED`	boolean	`false`	Enable semantic cache
`JORVIS_SEMANTIC_CACHE_TTL_MS`	number	`300000`	Cache TTL in milliseconds
`JORVIS_SEMANTIC_CACHE_MAX_ENTRIES`	number	`1000`	Max cache entries
`JORVIS_PRECISION_GUARD_ENABLED`	boolean	`false`	Enable precision guard (filter low-signal results)
`JORVIS_PRECISION_GUARD_MIN_K`	number	`5`	Minimum results to return
`JORVIS_PRECISION_GUARD_MAX_K`	number	`10`	Maximum results to return
`JORVIS_PRECISION_GUARD_MIN_SCORE`	number	`0.05`	Minimum rrfScore threshold
`JORVIS_VALUE_PROBING_ENABLED`	boolean	`false`	Enable column value probing for precision
`JORVIS_FORECAST_ENABLED`	boolean	`true`	Enable time-series forecasting (Task-019)
`FORECAST_SIDECAR_URL`	string	`http://localhost:8001`	URL for Python forecast sidecar
`RUN_E2E_TESTS`	boolean	`false`	Enable opt-in E2E/integration tests

Web Search Configuration

Variable	Type	Required	Description
`WEB_SEARCH_PROVIDER`	`tavily` \| `brave` \| `serper`	No	Search provider (default: `tavily`)
`TAVILY_API_KEY`	string	If provider=tavily	Tavily Search API key
`TAVILY_TIMEOUT_MS`	number	No	Tavily request timeout (default: 8000)
`BRAVE_API_KEY`	string	If provider=brave	Brave Search API key
`SERPER_API_KEY`	string	If provider=serper	Serper API key

Parallel Execution Limits

Configured in analytics-platform/src/ai/graph/parallel/parallel.constants.ts:

Constant	Value	Description
`MAX_PATHS`	3	Maximum parallel execution paths
`MAX_TOKENS_PER_PATH`	4000	Token limit per path
`MAX_TOTAL_TOKENS`	10000	Total token budget
`TIMEOUT_MS`	30000	Path execution timeout

Database Configuration

Variable	Pattern	Description
`JORVIS_DB_<PROFILE>_HOST`	string	Database host
`JORVIS_DB_<PROFILE>_PORT`	number	Database port
`JORVIS_DB_<PROFILE>_USER`	string	Database user
`JORVIS_DB_<PROFILE>_PASSWORD`	string	Database password
`JORVIS_DB_<PROFILE>_DATABASE`	string	Database name

LLM Configuration

Variable	Type	Description
`OPENAI_API_KEY`	string	OpenAI API key
`OPENAI_MODEL`	string	Model to use (default: gpt-4)

Open WebUI Native Images

These settings configure the canonical Jorvis image-generation surface through Open WebUI native Admin Panel -> Settings -> Images. They are PersistentConfig values in Open WebUI: new values apply automatically on a fresh bootstrap, but existing runtime state may continue using the saved Admin settings until an operator updates the Images panel or resets that persisted state.

Variable	Type	Default	Description
`ENABLE_IMAGE_GENERATION`	boolean	`true`	Enables the native Images surface in Open WebUI.
`USER_PERMISSIONS_FEATURES_IMAGE_GENERATION`	boolean	`true`	Allows end users to access the native Images surface.
`IMAGE_GENERATION_ENGINE`	string	`gemini`	Routes image creation through Gemini instead of a chat-model row.
`IMAGE_GENERATION_MODEL`	string	`models/gemini-3.1-flash-image-preview`	Canonical Gemini image-generation model ID for Jorvis.
`IMAGES_GEMINI_API_BASE_URL`	string	`https://generativelanguage.googleapis.com/v1beta`	Gemini image API base URL used by Open WebUI Images.
`IMAGES_GEMINI_ENDPOINT_METHOD`	string	`generateContent`	Required endpoint method for Gemini 3.1 Flash Image. Do not use Imagen-style `predict`.
`ENABLE_IMAGE_EDIT`	boolean	`true`	Keeps Open WebUI image editing aligned with the same native Gemini surface.
`IMAGE_EDIT_ENGINE`	string	`gemini`	Routes image editing through Gemini.
`IMAGE_EDIT_MODEL`	string	`models/gemini-3.1-flash-image-preview`	Canonical model ID for Gemini image-edit operations.
`IMAGES_EDIT_GEMINI_API_BASE_URL`	string	`https://generativelanguage.googleapis.com/v1beta`	Gemini image-edit API base URL used by Open WebUI Images.

Operational note:

Do not set ENABLE_PERSISTENT_CONFIG=false as the normal Jorvis path. That would force env precedence on every restart and make Admin changes ephemeral.
Do not expose Gemini image generation as a normal chat model in the picker.

Voice Gateway Configuration

Configured in analytics-platform/deploy/openwebui/docker-compose.yml:

Variable	Type	Required	Description
`EPHEMERAL_SECRET`	string	Yes	Random 32+ char string for session encryption
`GOOGLE_APPLICATION_CREDENTIALS`	path	Yes	Path to GCP service account JSON (e.g. `/secrets/sa.json`)
`VERTEX_PROJECT_ID`	string	Yes	GCP Project ID for Vertex AI
`STT_PROVIDER`	`gemini`	No	Speech-to-Text provider (default: gemini)
`STT_GEMINI_MODEL`	string	No	Gemini model for STT (default: gemini-2.5-flash)
`TTS_PROVIDER`	`gemini`	No	Text-to-Speech provider (default: gemini)
`TTS_DEFAULT_MODEL`	string	No	Gemini model for TTS (default: gemini-2.5-flash-preview-tts)
`GEMINI_LIVE_MODEL`	string	No	Gemini Live model (default: models/gemini-3.1-flash-live-preview)
`FALLBACK_GEMINI_LIVE_MODELS`	csv	No	Ordered Gemini Live fallback chain (default: `models/gemini-live-2.5-flash,models/gemini-live-2.5-pro,models/gemini-live-2.0-flash`)