Environment variables

The canonical template is .env.example at the repository root. After cloning:

bash

cp .env.example .env

Edit .env for your machine; never commit real secrets.

Docker Compose: defaults for Postgres and the app are set in docker-compose.yml. See Docker guide for container-specific usage.
Local pnpm dev: set DATABASE_URL, OLLAMA_URL (often http://localhost:11434), and WORKFLOW_LOCAL_DATA_DIR as described in .env.example.
Onboarding / first run: open /setup in the app. The wizard (DB-backed + provider steps) configures keys and creates the first admin; it does not write .env for you — keep using root .env.example as the template and sync values manually if needed.

Cross-links (optional): NUXT_PUBLIC_DOCS_SITE_URL points the Nuxt app header to your published docs (defaults to the GitHub Pages URL). For the VitePress build, VITEPRESS_DEMO_URL sets the hosted chat URL used in the nav and social icon; the default lives in docs/.vitepress/demo-app-url.ts and should match the Try demo hero link in docs/index.md. In GitHub Actions you can set repository variable VITEPRESS_DEMO_URL (see .github/workflows/pages.yml).

Provider selection

Two variables control which AI providers are active. If omitted, the runtime falls back to detecting the first present API key.

Variable	Values	Description
`EMBEDDING_PROVIDER`	`gemini` `openai` `voyage` `ollama-local`	Explicit embedding provider. Overrides key-presence detection.
`LLM_PROVIDER`	`anthropic` `openai` `mistral` `ollama-cloud` `ollama-local`	Explicit chat/LLM provider. Overrides key-presence detection.

Embedding providers

Variable	Provider	Description
`GOOGLE_API_KEY`	Gemini	API key from Google AI Studio. Uses `gemini-embedding-001` by default.
`OPENAI_API_KEY`	OpenAI	API key from OpenAI. Shared with OpenAI LLM if both are configured.
`VOYAGE_API_KEY`	Voyage AI	API key from Voyage AI. Uses `voyage-3` by default.
`OLLAMA_URL`	Ollama	Base URL, e.g. `http://localhost:11434`. Used for both embedding and LLM.
`OLLAMA_MODEL`	Ollama	Embedding model name (default: `nomic-embed-text`).
`EMBEDDING_MODEL`	all	Override the model for the active embedding provider.
`EMBEDDING_DIMENSIONS`	all	Vector size — must match the `pgvector` column (default: `768`).

Fallback order (when EMBEDDING_PROVIDER is not set): Gemini → OpenAI → Voyage → Ollama.

LLM / chat providers

Variable	Provider	Description
`ANTHROPIC_API_KEY`	Anthropic	API key from Anthropic console.
`ANTHROPIC_MODEL`	Anthropic	Model name (default: `claude-sonnet-4-6`).
`OPENAI_API_KEY`	OpenAI	Shared with embedding if both are configured.
`OPENAI_LLM_MODEL`	OpenAI	Chat model name (default: `gpt-4.1-mini`).
`MISTRAL_API_KEY`	Mistral	API key from Mistral console.
`MISTRAL_MODEL`	Mistral	Model name (default: `mistral-medium-latest`).
`OLLAMA_API_KEY`	Ollama Cloud	Auth key for Ollama Cloud.
`OLLAMA_LLM_MODEL`	Ollama	Chat model (default: `tinyllama` local / `kimi-k2.5:cloud` cloud).
`OLLAMA_CHAT_TIMEOUT_MS`	Ollama	LLM response timeout in ms (default: `180000`).
`OLLAMA_PLANNER_TIMEOUT_MS`	Ollama	Planner/tool step timeout in ms (default: `60000`).

Fallback order (when LLM_PROVIDER is not set): Anthropic → Mistral → OpenAI → Ollama.

Database

Variable	Description
`DATABASE_URL`	Full Postgres connection string. Required.

Self-hosted pgvector example:

DATABASE_URL=postgresql://postgres:password@localhost:5432/rag_db

Supabase Vector example (direct):

DATABASE_URL=postgresql://postgres:password@db.<project-ref>.supabase.co:5432/postgres

Supabase Vector example (transaction pooler — recommended for serverless):

DATABASE_URL=postgresql://postgres.<project-ref>:password@aws-0-<region>.pooler.supabase.com:6543/postgres

The setup wizard (step 2) can help you pick a Postgres connection style; still copy the final string into .env yourself.

Application

Variable	Default	Description
`MEMORY_SCOPE`	`local_per_user`	`local_per_user` / `global` / `disabled`
`MEMORY_PROACTIVE`	`true`	Auto-save user facts to memory
`ADMIN_API_KEY`	—	Optional key to protect admin endpoints
`WORKFLOW_LOCAL_DATA_DIR`	`./data/workflow`	Durable workflow state directory

For runtime-configurable values stored in the database (model, chunking, search, RAG tunables) see Settings. Anything in this file is read once at boot from the environment.

Environment variables ​

Provider selection ​

Embedding providers ​

LLM / chat providers ​