Requirements

To use the AI Assistant, you need an account and API key from one of the supported providers:

  • OpenAI - GPT-4o, GPT-4, o1, o3, gpt-5 and other models
  • Gemini (Google) - gemini-2.0-flash, gemini-1.5-pro and other models
  • Claude (Anthropic) - claude-opus-4, claude-sonnet-4 and other models
  • Grok (xAI) - grok-3, grok-2 and other models

Step-by-step AI configuration

1. Open the settings

Click the gear icon (⚙) in the top-left corner → AI tab.

2. Select a provider

Select one of the available providers from the “AI provider” list.

3. Enter the API key

Paste the API key obtained from the provider. The key is encrypted and is never displayed in full after it has been saved.

Where can I get an API key?

  • OpenAI: https://platform.openai.com/api-keys
  • Gemini: https://aistudio.google.com/apikey
  • Claude: https://console.anthropic.com/settings/keys
  • Grok: https://console.x.ai/

4. Select a model

After entering the API key, click the refresh button (↻) next to the model list - the application will retrieve the current list of available models from the provider. Select a model from the list.

Models marked (recommended) are proven, reliable choices.

5. Save the configuration

Click “Save AI settings”. The application will test the connection to the model and display any warnings.

Advanced settings

In the “AI settings” section, several options are grouped under the expandable Advanced subsection (click the heading to expand it):

  • Maximum response length - controls the length of AI responses (1024-16384 tokens). A higher value means longer responses, but higher costs. Default: 4096.
  • Save AI Assistant conversation history - the administrator can enable or disable saving conversation history in the database. Enabled by default.
  • AI attachment summaries - AI creates short file summaries to respond faster in tasks with many attachments. The default mode is Manual (summaries are only created after clicking Create summary next to the section name - there is no automatic API usage). The list of skipped/failed files is available under “Details”.
  • Semantic search - searches by meaning rather than keywords alone. Requires OpenAI or Gemini as the provider (enabled by default for these providers).
  • Allow Mail to be used as an AI knowledge source - disabled by default. When enabled, each user can select their own mailbox or email thread as a source for a conversation with the Assistant (no one, including the administrator, can see anyone else’s mail). The content of the selected source is then sent to the external AI provider, and correspondence contains third-party data - which is why the administrator’s informed consent is required. Mail is never added to the knowledge base indexed in the background; disabling this setting also disconnects any mail sources selected previously.
  • Daily limit for AI requests from Automations - the “AI evaluates” condition and “write with AI” action query the model for every matching event; the limit (default: 200/day) protects your budget with the provider. Today’s usage is shown next to the field. Enter 0 to disable the limit. Once the limit is reached, Recipes using AI are skipped for the rest of the day and an entry is added to the history - the counter resets at midnight and everything resumes automatically.

AI knowledge base

The “AI knowledge base” section in the AI settings (administrators only) allows you to manage the Assistant’s automatic context.

Automatic knowledge base updates

The Automatic knowledge base updates switch (Yes / No) controls whether the knowledge base is refreshed automatically:

  • Yes (default) - the knowledge base is updated automatically once a day in the background. Users can continue using the application as normal.
  • No - automatic updates are disabled. The administrator updates the knowledge base manually using the Update AI knowledge base button when needed. Recommended when there are very large numbers of tasks or chats.

Manual update

The Update AI knowledge base button immediately processes all tasks and chats that do not yet have a summary or have been changed since the last update. It works regardless of the automatic update setting.

What is included in the knowledge base

  • Tasks and chats - summarised by the model (decisions, conclusions and agreements).
  • Notes and Files (since 2.2.273) - added with their original content, divided into sections. This allows the Assistant to find answers in company documents even if no one selects them as a source - simply ask “what is the fuel limit?”. Indexing documents does not use model requests (they are not summarised), so it costs nothing and works even when AI is disabled - in that case, keyword search is used. The following text file formats can be read: TXT, CSV, Markdown, DOCX, XLSX and PDFs with a text layer (scans without text are skipped).

Privacy in the knowledge base

Notes and Files retain their own permissions: belonging to a Workspace does not grant access to them. The Assistant’s responses will only include documents that you can already open - your own documents and those shared with you. This also applies to administrators. A deleted document (bin / trash) is removed from the knowledge base during the next update.

Troubleshooting

“Error: invalid API key”

Check that the pasted key is correct. Make sure your account with the provider has active billing/limits.

“The model does not support chat mode”

The selected model is not a chat model. Return to the AI settings and select a different model from the list.

“Request limit exceeded (rate limit)”

Your account with the provider has reached its request limit. Wait a moment or upgrade to a higher plan.

The model list is empty

Click the refresh button (↻) next to the model list. If it is still empty - check the API key and your internet connection.