AI settings & oversight

Turn the assistant on with your own key, set budgets, manage skills and review conversations.

Turning it on

SWARECO AI is off by default. An admin or above turns it on in Settings → Account settings → AI assistant:

1. Switch the assistant on.
2. Choose the provider — Anthropic today.
3. Paste your account's API key.
4. Choose the model and how much thinking effort it should use.

The key is your account's own, so the provider bills your account directly for model use. It is encrypted at rest and is write-only: the page shows only its last four characters, and a blank field keeps the stored key. Test key checks that it works without running a chat, and Remove key clears it. Changing the provider clears the key, model and effort.

While it is off, every part of the assistant stops and no request is sent to the provider. The button in the top bar still opens a panel that says so.

Budgets

Set a monthly budget in dollars for the account and, optionally, for each member. The limit is checked before a message is accepted and again before every model call, so several chats running at once cannot overspend together. The settings page shows Spent this month.

Costs are estimates from the model's list price. For a hard ceiling, also set a spend limit on the key with your provider.

Skills

Admins and owners manage skills in Settings → AI skills. A skill has a name, a command (the /word people type), a one-line description, the instructions, an on/off switch and an optional minimum role. The four defaults — Proposal, Follow-up email, Deal summary and Call notes — are added the first time the assistant is switched on, and a default you delete stays deleted.

A skill is plain text sent to the assistant as part of the request. It is never run as code.

AI activity

Settings → AI activity lists every SWARECO AI conversation in the account, for admins and owners. You see the chats of people at or below your own role: an admin reviews members' and admins' chats, an owner reviews everyone's. It keeps working while the assistant is switched off, so past conversations stay reviewable.

• The list shows who, which conversation, which record, when it last moved, and per chat the number of messages, tool calls, changes applied and the estimated cost. Filter by person, by date range, or to chats that changed data.
• A transcript opens read-only: what the person saw, then the lookups the assistant made and the changes it applied, with their status, who decided, and the before and after values.
• Exports — the filtered list as CSV (the 5,000 most recent chats, with facts about each chat but not message text) and a single transcript as JSON.

Retention

In the same AI assistant settings, choose how long to keep conversations: forever (the default), or 30, 90 or 365 days. A nightly job removes older conversations along with their messages, lookups, usage records and proposed changes. A conversation that was active this month is never removed, because the month's spend is worked out from them. Your contacts, deals and their history are untouched.

Anyone can archive their own chats to tidy their list. Archived chats stay in the activity log until retention removes them. There is no way for a user to delete a chat.

What keeps it inside your permissions

• The assistant reads and writes through the same permission checks as the rest of the app, as the person who is chatting. It sees what they see and can propose only what their role allows.
• Every lookup it makes is recorded, with who asked, the tool, the arguments and the outcome.
• Prompts and replies are not sent to our error tracking.
• Because every change waits for a click, text planted in a record — including text typed into a public form — cannot make the assistant change anything on its own.

The same tools are available to outside assistants over MCP.

Something inaccurate or missing? Tell us — we would rather fix the docs than have you guess.