AI providers

Bring your own key

Claril's AI layer is brand-agnostic and BYOK. No keys ship with the product, and AI is never required — the deterministic inspector is fully useful with nothing connected.

AI is progressive enhancement. A quiet AI: off / connected pill reflects status, and marks features AI makes better — never features it blocks.

Supported providers

Connect any combination of these six providers in your personal space or in an organization:

  • Anthropic

    Claude models.

  • OpenAI

    GPT models.

  • Google

    Gemini models.

  • Mistral

    Mistral hosted models.

  • Ollama

    Local / self-hosted inference — no key required, just a reachable endpoint.

  • OpenRouter

    An OpenAI-compatible gateway fronting many providers; "auto" routes for you.

Personal vs organization

Where you connect a provider matches where you're working (switch with the context switcher):

  • Personal space.Your own keys power AI in your solo projects. They're scoped to you and never shared.
  • Organization. Owners and admins connect shared keys the whole org can use, alongside the shared Asset Catalog. Each org connects each provider at most once.

How keys are stored

  • Keys are encrypted at rest with AES-256-GCM, using CLARIL_ENCRYPTION_KEY (falling back to BETTER_AUTH_SECRETif it's not set). They are never returned to the client and never baked into any build.
  • Keys are strictly scoped to the personal account or organization that added them.
  • On a self-hosted instance, inference requests go from your server straight to your chosen provider with your credentials — no middleman.
  • Ollama needs no key; provide a base URL for your local or self-hosted endpoint instead.

Default model & per-session switching

Each space picks one default modelacross its connected providers — the model the advisor uses unless told otherwise. There's a personal default for your solo work and an org default for organization work; exactly one default exists per space at a time.

  1. Each connected provider has its own default model (seeded from a sensible recommendation, changeable in settings).
  2. The space's default model selects which connected (provider, model) the advisor uses by default. Connecting your first or only provider auto-sets the default for you.
  3. In the workbench, a compact selector lists every model across all connected providers so you can switch the model per session — and optionally set your choice as the new default.

Connecting a provider

  1. Open settings for your current space (personal settings, or organization settings) and find the AI providers / connections section.
  2. Add a provider, paste your API key (and an optional base URL for proxies or self-hosted endpoints), and choose a default model.
  3. Use Test to verify the credentials reach the provider. Your first connection becomes the default automatically; adjust it any time.

In an organization, these controls require an owner or admin; in your personal space they're always yours. Once a usable connection exists, the AI co-editor — generate BPMN from a prompt, chat grounded on findings and the catalog, AI-proposed edits you review, and Markdown doc-gen — becomes available. See Getting started for the end-to-end flow.