Asking a Bot
Part of the MountDev Ops Bots documentation. Updated November 20, 2026
MountDev Ops Bots documentation | Updated November 20, 2026
A person asks a bot a question and gets an answer written on your company's own Claude key (action bots.ask). The bot only writes an answer: it sends nothing, changes nothing, and never reads a record in this update.
Our Commitment
MountDev Ops is built security first. Your business data belongs to you, and you can export all of it at any time. We never use analytics or tracking of any kind inside MountDev Ops.
Every update passes all 8,865 automated checks before it ships, and every screen change is walked through before release.
What Has to Be True
Every one of these is checked on every question, by the server, and never by the bot's instructions. If one is not true, nothing is asked of Claude and the message says which one.
- The bot is on. A person turns it on, and only once its three limits are typed in (see Spending Limits).
- The place is switched on for the bot (Where You Can Talk to It). The API and MCP and a website chat are Planned, and a bot refuses them.
- You are one of the people the bot answers (Who Can Talk to It): everyone, chosen roles, chosen teams, or chosen people.
- You have the right to ask a bot in Settings, Permissions. Owners, admins, and members have it to start.
- Your company has a Claude API key in Settings, AI Agents. There is no shared key and no fallback: with no key of your own the bot does not run, and the message says to add one.
- The person who made the bot is still on the team and still has Bots. If not, the bot is paused, owners and admins are told once, and it stays paused until a person with Bots turns it on again.
Claude is the only provider that runs for now. A bot set to OpenAI does not run and the message says to choose Claude in Who It Is (OpenAI is Planned).
Where You Ask It
Asking in Team Chat, Project Chat, Helpdesk tickets, and Workflows arrives in later updates (Planned). Today you try a bot in Test.
What the Bot Uses
- Its instructions, tone, language, and label.
- What it learns from, only when approved: pasted text, and an outside link's text once you approved its summary. A refreshed page is not used until you approve its new summary, and the old approved text stays in use meanwhile. Removed items are not used. Docs and website pages, links to items in your workspace, and your workspace data are read when asking in a later update (Planned).
- What the person typed, and any material they pasted.
Anything pasted as material, and every piece of learned text, is handed to Claude as material to read, never as instructions. A request hidden inside an email, a web page, or a file is not followed.
A bot never shows anyone more than that person could see. In this update a bot reads no records at all, so it can show only its approved learning and the words in the question.
What the Bot Never Does
It never sends email, changes a record, spends money, changes people or access, or takes any action. It writes an answer for a person to read. Its label (Bot, Automated, or Virtual Assistant) is always shown with its name.
How a Run Ends
Each run has one result, shown in Activity:
- Answered: the answer came back within the limits.
- Reached Its Limit: the answer came back but used up the limit for one run. The bot stops there and says what you wrote in its Handoff words, if its hand-off includes its limit.
- Stopped: the monthly limit was reached, or the question was too large for the limit for one run. Nothing was asked of Claude.
- Paused: the person who made the bot is gone.
- Could Not Answer: Claude was busy, did not accept your key, could not be reached, declined, or sent nothing back. The message says which, in plain words. A key you pasted is never shown.
What Is Written Down
Each run is one line in the Activity Log with the bot's name and label, the person who made it, where it was asked, the number of steps, and the cost, for example "Penny (Bot), made by Dana Lee: answered (asked in Team Chat, 1 step, $0.0040)". The person who asked is logged with it. The line and the run's row hold counts only: never the question, the answer, a name, or an email. A run takes one step in this update, because a bot takes no actions yet.
