AI Settings, models and limits
Choose the engine and model, save keys, set live voice, decide which fields are never sent, and set limits.
Required permission: ai.configure
Before you begin
You need ai.configure. Open AI Assistant > Settings. Settings have six tabs: General, Models, Live voice, Security & privacy, Limits & retention, and the Activity log and Usage tabs for users with ai.activity (see Activity log, usage and switches).
Use test keys while you configure. Keys are stored sealed. They are never shown again; the screen shows 'A key is saved - type to replace it'.
Steps
General
- Open the General tab.
- Tick or clear AI enabled. When cleared, new chats and messages are refused.
- Tick or clear the switches for Voice, File upload, Knowledge, MCP and Write actions enabled.
- Choose the Default agent and the Assistant response language (Same as the question, English, Arabic, Hindi, Malayalam or Urdu).
- Edit the Quick prompts, one per line. Up to 12 are kept, each cut to 80 characters.
- Select Save.
Models
- Open Models. Choose the Engine: a2N built-in (no model, no key) or Anthropic Claude.
- Choose Token saving: Smart (answer common questions without the model) or Always ask the model.
- Choose Effort: Low (fastest), Medium or High (most thorough).
- Choose the Default model and the Fast model from the list.
- Paste the Anthropic API key and select Save, then Test connection. A good result reads 'claude-opus-5 answered: ready'. Remove key switches the engine back to the built-in one.
Live voice
- Open Live voice. Tick Live voice enabled, choose the Live model, the Voice (Kore, Puck, Charon, Fenrir, Aoede, Leda, Orus or Zephyr) and paste the Google Gemini API key.
- Select Test live connection. A good result reads 'gemini-3.8-live is ready with the Kore voice.'
Security and privacy
- Open Security & privacy. Under Never send these fields to AI, list one field key per line, such as
phone,emailortax_id(up to 100). - Select Save.
Limits and retention
- Open Limits & retention. Set the values and save:
| Setting | Allowed range | Default |
|---|---|---|
| Messages per minute (per user) | 1 to 120 | 20 |
| Daily action limit | 1 to 10,000 | 200 |
| Max file size (MB) | 1 to 50 | 10 |
| Max spreadsheet rows | 100 to 100,000 | 5,000 |
| Chat retention (days) | 7 to 3,650 | 365 |
| File retention (days) | 1 to 3,650 | 90 |
What happens next
- A saved setting takes effect on the next message.
- Fields in the never-send list are removed from what the model, live voice and the activity log input receive. The on-screen table still shows them if your role may see them.
Good to know
- 'Add an API key to use Claude, or choose the built-in engine.' appears when Claude is chosen without a key. 'The API key was refused.' means the key is wrong. 'Add a Google Gemini API key first.' appears for live voice.
- Out-of-range values read, for example, 'Enter a number from 1 to 120.' and 'Enter a whole number.'
- 'Choose a model from the list.' 'Choose a live model from the list.' 'Choose a voice from the list.'
- Two administrators saving at once: 'Somebody else changed these settings. Reload and try again.'
- Live voice works only when AI, Voice and Live voice are all on and a Google key is saved. Without it: 'Live voice is not set up. An administrator adds a Google Gemini key in AI Settings > Live voice.'
- The daily action limit, chat retention and file retention are stored but are not enforced yet. There is no cleanup job and no token budget; Usage reports an estimate only.
- The Fast model is saved but is not used yet.
- Live voice is not subject to the messages-per-minute limit.
- Haiku 4.5 does not receive the Effort setting.