Read this page before you let staff use the AI Assistant. It says what an agent can read, what leaves your account, who sees the chats and what limits apply. The limits are settings of your installation, so the page describes how they behave and gives no numbers.
An agent reads live data for the accounts and the modules your role can open. The chat says so: "Agents read live data for the accounts and modules your role can access."
- Every lookup runs as you. A role without Reports cannot get a report through an agent.
- Agents mostly read. The few changes they can make are listed in What each agent can do.
- If the app cannot check your permissions at that moment, the agent refuses with "Could not verify your permissions right now. Please try again."
Your question, and the data the agent looked up to answer it, go to the AI provider that serves your organisation. That is a service such as OpenAI, Anthropic, Google Gemini, or a server that speaks the OpenAI format, including one the provider owner runs. Which one depends on the provider set up for your account, or granted to it by an account above.
- The provider's API key belongs to the account that added it under LLM Providers. The app stores it encrypted, never shows it again, and a sub-account that uses it never sees it.
- A provider reached over plain http sends the key and the conversation unencrypted. The provider form warns about this.
Ask your administrator which provider serves your organisation and under which terms before you put sensitive questions to the assistant.
Your history is private to you. The chat shows "History is private to you", and the Recent conversations list on the AI Agents screen shows only your own. You can export a conversation or delete it. Deleting is permanent: the app removes the conversation and its messages and cannot bring them back.
The AI settings of an account show usage totals: the credits used, the number of messages and tokens, and the split by agent.
Each organisation has a monthly allowance of AI credits. Every answer uses some of them, depending on how much text goes back and forth.
- The allowance is the Monthly credit limit that an account above set for yours in its AI Assistant section. Left blank, your organisation has no limit of its own, and the pool cap of the provider key it uses applies. A key saved without a pool cap is held to the platform default. An administrator can add credits for the current month only.
- The meter on the AI Agents screen shows the credits used and the limit. At a warning level your platform sets, it says "You have used N% of this month's AI credits."
- At the limit, the meter turns red and chats answer with "Your organisation has used all N of its AI credits for this month." The allowance starts again at the beginning of the next month, or an administrator can raise it.
- Every provider key has its own monthly pool cap, shared by every account billed to its key, including the account that owns the key. When it runs out, chats say "The AI provider key your organisation uses has reached its monthly credit pool. Ask the account that granted you AI to raise it."
The credits belong to the organisation, not to one user. One heavy user can use up the allowance for everyone.
¶ Message and conversation length
- A message has a length limit. If you go over it, the app says "That message is too long. Keep it under N characters."
- A chat holds a limited number of messages. When it is nearly full, the chat says "This chat is nearly full" and how many messages are left. When it is full, the composer says "This chat is full. Start a new chat to carry on." Select New chat. Answers stay faster and more accurate in a fresh chat.
- An agent stops after a set number of lookups for one question. If it runs out, ask a narrower question, for example one vehicle or one week.
The assistant writes plausible text from the data it read. It can misread a question, pick the wrong period or summarise a long list badly. For anything you act on, such as a service decision, a billing check or a disciplinary matter, confirm the numbers on the report or unit screen the answer is based on. In a chat, the activity rows of an answer list the lookups the agent made: Model arguments, Service request and Service response.
The switch is in the account's form, in the AI Assistant section. Users with Manage AI Assistant Settings on the account above can change it.
- AI Assistant enabled shows the menu entry for the organisation. Turning it off also turns AI off for every account below it.
- Monthly credit limit, LLM provider and Model say how it runs. A limit is 1 credit or more, or blank for none. Model offers the provider's default model by name. If the provider has no default model, you must choose a model for the account. See LLM Providers.
- Allow this account to grant AI to its sub-accounts lets a dealer pass the provider on, within its own limit.
- The section shows the state of the account: Enabled, Disabled here, Disabled by a parent account, "Suspended" when the platform has suspended AI for the account, or Not granted.
- An account with its own enabled default provider runs AI on that key, even when the account above has switched its grant off. The state then reads "Enabled — own provider". See LLM provider form.
When AI is off, users see "The AI Assistant is switched off for your organisation." The menu entry disappears once the app knows.