Platform

Configuration

Start with the root .env.example. This page summarizes selected settings; the example file and each chart's values are the sources for the complete configuration.

apps/console and apps/siteare separate frontends. The public website needs no authentication or backend configuration. Services consume different subsets of environment settings. The local make dev-all target checks database, Supabase, and signing values; console production builds also need public API and documentation URLs.

Core runtime

Variable (local startup requirements marked)	Default	Description
`ENVIRONMENT`	`development`	Environment label used across logs and startup behaviour.
`DEBUG`	`true`	Enables debug-oriented behaviour in local development.
`LOG_LEVEL`	`DEBUG`	Application log level.
`LOG_FORMAT`	`text`	Log formatter. Local development uses plain text by default.
`SECRET_KEY`req	`change-this-in-production`	JWT signing key. Replace the example value before deployment.
`FIELD_ENCRYPTION_KEY`	`empty`	Fernet key for encrypting stored secrets and other sensitive fields. Local setup can generate one.
`ACCESS_TOKEN_EXPIRE_MINUTES`	`30`	Access token lifetime.
`REFRESH_TOKEN_EXPIRE_DAYS`	`7`	Refresh token lifetime.

Ports and browser URLs

Variable (local startup requirements marked)	Default	Description
`API_PORT`	`8000`	Local FastAPI port.
`CONSOLE_PORT`	`3000`	Local port for apps/console.
`SITE_PORT`	`3001`	Local port for apps/site.
`WORKER_PORT`	`5000`	Legacy Makefile value used in its printed worker health URL; the worker health server uses WORKER_HEALTH_PORT (default 8081).
`DASHBOARD_PORT`	`8080`	Taskiq dashboard port.
`NEXT_PUBLIC_SITE_URL`	`http://localhost:3001`	Site URL used by the console; set for production builds.
`SITE_URL`	`http://localhost:3000`	Console URL used by local Supabase Auth redirect configuration.
`NEXT_PUBLIC_API_URL`	`http://localhost:8000`	Browser-facing API URL; set for production builds.

Database and Supabase

Variable (local startup requirements marked)	Default	Description
`DATABASE_URL`req	`placeholder in .env.example`	Primary superuser connection used by migrations and seed scripts.
`DATABASE_URL_APP`req	`placeholder in .env.example`	Runtime database connection for the API role.
`DATABASE_URL_WORKER`req	`placeholder in .env.example`	Runtime database connection for the worker role.
`DATABASE_POOL_SIZE`	`20`	Async SQLAlchemy pool size.
`DATABASE_MAX_OVERFLOW`	`10`	Allowed burst connections above the base pool size.
`SUPABASE_URL`req	`placeholder in .env.example`	Supabase project URL used by the configured auth provider.
`SUPABASE_ANON_KEY`req	`placeholder in .env.example`	Public Supabase key used by browser auth.
`SUPABASE_SERVICE_ROLE_KEY`req	`placeholder in .env.example`	Privileged key used by setup and seed scripts. Keep it server-side.

Redis and worker control

Variable (local startup requirements marked)	Default	Description
`REDIS_URL`	`redis://localhost:6379/0`	Primary Redis connection string.
`REDIS_MAX_CONNECTIONS`	`10`	Redis client connection cap.
`REDIS_PASSWORD`	`empty`	Production password injected into REDIS_URL at startup.
`TASKIQ_DASHBOARD_URL`	`http://localhost:8080`	Worker event dashboard endpoint.
`TASKIQ_DASHBOARD_TOKEN`	`empty`	Dashboard auth token.
`EMBEDDING_MAX_CONCURRENCY`	`2`	Caps concurrent embedding work per worker process.
`EMBEDDING_TIMEOUT_SECONDS`	`30.0`	Timeout for one embedding batch.
`WORKER_MAX_ASYNC_TASKS`	`8`	Taskiq worker concurrency ceiling.
`WORKER_MAX_PREFETCH`	`8`	How many jobs a worker pulls ahead from Redis.

LLM and retrieval

Variable (local startup requirements marked)	Default	Description
`LLM_MODEL`	`anthropic/claude-3-haiku-20240307`	Model routing target shown in the checked-in .env.example.
`LLM_MAX_TOKENS`	`1024`	Default completion token budget.
`OLLAMA_API_BASE`	`http://localhost:11434`	Base URL for local or internal Ollama inference.
`ANTHROPIC_API_KEY`	`empty`	Required when routing to anthropic/* models.
`OPENAI_API_KEY`	`empty`	Required when routing to openai/* models.
`GEMINI_API_KEY`	`empty`	Required when routing to gemini/* models.
`GROQ_API_KEY`	`empty`	Required when routing to groq/* models.
`EMBEDDING_MODEL`	`local/all-MiniLM-L6-v2`	provider/model for chunk and query vectors (local/... or voyage/...).
`RERANKER_MODEL`	`local/cross-encoder/ms-marco-MiniLM-L-6-v2`	provider/model for reranking. Empty disables reranking.
`EMBEDDING_API_KEY`	`empty`	Required for hosted embedding or rerank providers (voyage/*).
`PARSER_MODEL`	`local/docling`	Provider/name for document parsing: local/docling or llamaparse/<tier>. Managed parsing sends documents to the provider.
`PARSER_API_KEY`	`empty`	Required for managed parsers (llamaparse/*).
`CHUNK_SIZE`	`512`	Default chunk length for ingestion.
`CHUNK_OVERLAP`	`50`	Token overlap between adjacent chunks.
`RETRIEVAL_TOP_K`	`5`	How many retrieved chunks are fed into answer generation.
`EMBEDDING_DIMENSION`	`384`	Output size of the embedding model. Startup fails if it does not match the stored vectors.
`MCP_TIMEOUT_SECONDS`	`30`	Timeout for MCP tool execution.
`MCP_MAX_RETRIES`	`3`	Retry cap for MCP tool execution.

Feature flags and rate limits

Variable (local startup requirements marked)	Default	Description
`ENABLE_RAG`	`true`	Master switch for retrieval and answer generation features.
`ENABLE_MCP`	`true`	Enables MCP capability plumbing.
`ENABLE_AUDIT_LOGGING`	`true`	Turns on audit logging.
`REQUIRE_APPROVAL_FOR_WRITES`	`true`	Forces HITL approval before write actions execute.
`RATE_LIMIT_ENABLED`	`true`	Master switch for API rate limiting.
`RATE_LIMIT_AUTH_PER_MINUTE`	`5`	Auth request cap per IP per minute.
`RATE_LIMIT_QUERY_PER_MINUTE`	`60`	Query request cap per IP per minute.
`RATE_LIMIT_UPLOAD_PER_HOUR`	`20`	Document upload cap per IP per hour.
`RATE_LIMIT_CONNECTOR_PER_HOUR`	`10`	Connector sync cap per IP per hour.

Model examples

Examples below are provider/model identifiers from .env.example. Set the matching provider key. Hosted model, embedding, and parser requests send data to those providers.

anthropic/claude-3-haiku-20240307Example configured in .env.example

anthropic/claude-sonnet-4-6Anthropic model example

openai/gpt-4oOpenAI model example

gemini/gemini-2.0-flashGemini model example

groq/llama-3.3-70b-versatileGroq model example

ollama/llama3.1Ollama example; requires a reachable service and pulled model