AI in a tenant
Managed AI or your own key, monthly credits, choosing a model, and standing instructions.
The assistant (in Chat and in the sidebar) and automations run on the tenant's AI source, which a TenantAdmin manages under Settings → Tenant → AI. There are two sources: managed AI, where Beetl provides the models from monthly plan credits, and bring your own key. If the tenant has registered a key, Beetl uses it; otherwise it uses managed AI.
Managed AI
Managed AI needs no setup, so a new tenant can chat and run automations from the start. You can choose from these models:
- Gemini 3.5 Flash Lite, the default
- Claude Sonnet 5.5
- Claude Opus 5.5
- GPT-6.1 Sol
- Mistral Medium 3.5
The tenant's plan sets its monthly credits. The AI tab shows the credits remaining, and the allowance resets on the 1st of each month. Larger models and higher reasoning effort use credits faster. Chat titles and generated pipeline docs do not use credits. To change plans, contact Beetl.
When credits run out, chat turns and automation runs fail with "You have reached your monthly AI limit". Admins are offered the option to bring their own key; others are told to ask an admin. A message refused before its turn starts goes back into the composer, so nothing you typed is lost.
Bring your own key
Choose Bring your own key on the AI tab:
| Field | What to enter |
|---|---|
| Provider | Gemini, OpenRouter or OpenAI-compatible |
| Base URL | OpenAI-compatible only, for example https://api.openai.com/v1 |
| API key | Your provider key |
| Default model | Used by automations and new chats. Empty means the first chat model the key offers. |
| Models in chat | The extra models the chat picker offers. None checked means all of them. |
The key is stored encrypted and never shown again. To rotate it, paste a new key and choose Replace. Your provider bills these calls, and Beetl's credits don't apply. To go back, choose Managed and then Remove key and use managed AI.
Choosing a model and effort
The chat composer and the automation editor each have a model chip and a reasoning-effort chip (Default, Low, Medium or High). Each chat session and each automation keeps its own choice.
Assistant instructions
The AI tab also holds up to 4,000 characters of workspace conventions, such as which dataset is the source of truth for customers or how your team names pipelines. Every chat turn and automation run receives them, and the assistant treats them as trusted guidance, so write only what you want followed. Only TenantAdmins can read or change them. MCP clients don't receive them, because they bring their own model.
Where model traffic goes
The model receives your messages, the assistant instructions, and whatever the assistant reads through its tools, including query rows and pipeline definitions. With your own key, that goes to the provider you registered. For managed AI, see Where your data lives.
Related
Last updated on