AI in your plan, and keeping data from AI

Your base's plan includes the AI that the base uses, for the normal use of a base (Upgrade to the Pro plan). The plan's AI pays for:

• the assistant: its answers, the changes it makes, the code it writes included, its summaries, and the titles, summaries and topics of the parts of your conversation;
• the AI menu beside a comment, a mail's answer and a long text field (Write, rewrite and translate with AI);
• the agents on Base Captain, in a chat with you too;
• the workflows' steps and the automations' runs;
• Triage, which reads the work that comes in, starts the right workflow and fills in or suggests values;
• a question about a web page or a page's picture by the assistant or an agent, a look at a web page by a machine agent, and pi when it takes its model from Base Captain;
• a web search or a page read through Base Captain's AnyAPI account, when no machine of yours answers it and the base has no AnyAPI key of its own;
• reading a file, a picture or a PDF.

Some agents do not use the plan's AI:

• A machine agent on its own Claude Code or Codex login does not use the plan's AI, and goes on when the plan's AI stops (Run agents on your own computer: machines).
• An agent on the Claude API or the OpenAI API does not use it either: your base's account at Anthropic or OpenAI pays for its AI, on every plan, and the agent goes on when the plan's AI stops. Only a web search or a page read through Base Captain's AnyAPI account in its runs uses the plan's AI (Run an agent on your own Claude API or OpenAI API key).

Fair-use limits on the AI protect Base Captain's service for all customers (Terms of service). When a base reaches one, the AI pauses until the month ends, and a sentence says what stopped and when it works again:

• Agents and workflows stop first, so you can still change your base with the assistant. A workflow's items wait where they are, and go on when the AI works again. A chat with an agent says "The AI has reached this base's fair-use limit for agents and workflows this month. They work again on 1 November. A machine agent on your own subscription goes on at no cost. Contact us if you need more."
• The assistant stops later: "The AI has reached this base's fair-use limit for this month. It works again on 1 November. A machine agent on your own subscription goes on at no cost. Contact us if you need more."
• Back on the 1st of the next month, at midnight UTC. Where midnight UTC is another hour of your day, the panel names it.
• Told: when the AI stops, the base's owners and its builders get one email with the same sentence, once a month.

Each person also has fair-use limits of their own on the assistant, across all their bases:

• Bursts: after many messages in a short time, the panel says when to send the next one: "You sent many messages to the assistant in a short time. It works again for you in 40 seconds. Contact us if you need more." Each click in the AI menu counts as a message.
• A day: past a person's day of use, the panel says "You reached your fair-use limit of the assistant for today. It works again for you tomorrow at 00:00 UTC. Contact us if you need more."

Limits and details:

• A message refused by a limit is not kept: send it again once the assistant is back.
• An answer already under way when a limit is reached finishes.
• A card waiting for you is always answered: Apply and Cancel still work, and the assistant stops there instead of going on.

The Storage page shows your base's Storage to the admins: click Activity, then Storage. It reads such as "620 MB of 10 GB": what the base keeps, its records, the Files page and Chat included, against its plan's storage and the storage an owner added, counted every hour. Past it, new items and uploads are refused (What happens past a limit).

The base's Settings, Plan shows the same to its owners (Upgrade to the Pro plan). An admin's API token, the assistant and an AI client read the same number (Use your base from a calendar or a script).

The Statistics page shows what the base did over its whole life, to the users who build the base: click Activity, then Statistics. It counts:

• for each table, the items added, changed and deleted, and how many it holds; you can see them split by a person or by what made them, such as the assistant, an automation, a workflow or an import;
• for each automation, how many times it fired, went through and failed;
• for each workflow, the items it started, finished, failed and stopped;
• the agents' runs, the jobs, the imports, the exports and the mails, each done and failed.

Pick Hour, Day, Week or Month and a range of dates, from the base's first day to today. Days, weeks and months are cut in your own time zone, and weeks start on your own first day of the week: Settings, Appearance, Time zone and Start week on. Two users in two time zones each see their own "last week".

Limits and details:

• The number of items of a period is the one at its end.
• In a time zone a half or three quarters of an hour from UTC, a day starts that much off.
• The numbers are counted once an hour, so the current hour shows after it ends.
• The numbers are kept for good: a renamed table or automation keeps its past, and a deleted one's numbers stay.
• A failure is only a count; why it failed is not shown.
• The admins and the builders see the statistics; the storage and the AI stay the admins'.

Your API token, the assistant and AI clients read the same numbers (Use your base from a calendar or a script).

Keep item data from the assistant and agents, a switch in the base's settings, keeps your items, the messages of Chat and the files of Files from the assistant, the agents and AI clients. Only the base's admins change it.

1. Click Settings, then AI settings.
2. Switch on Keep item data from the assistant and agents.
3. Click Save settings.

Switch it off the same way. While the switch is on:

• The assistant still changes the base for its builders, its tables, views and settings, but it is refused your items, the messages of Chat and the files of Files: it cannot read, count or change them, and says so.
• The base's agents and Triage do not run, and a mention of the base's assistant or one of its agents is refused under its message (Ask an agent in a channel).
• The AI menu beside comments, mail answers and long text fields does not show (Write, rewrite and translate with AI).
• Base Captain's MCP server refuses every call of every AI client in the base, whatever its token, the Claude app and ChatGPT included: "This base keeps its data from the assistant and agents." (What your AI can do in Base Captain). Your other bases still answer it.
• A script with an API token is not AI, so its own calls still work; the assistant and the agents that the script chats with cannot reach the base's data, as in the browser.

Only an admin changes the switch, in the base. No assistant, agent or AI client changes it, and a change one of them asks for is refused: "Only an admin can change this, in Settings, AI settings."