Ask My AI
Part of the MountDev Ops Workflows documentation. Updated November 14, 2026
MountDev Ops Workflows documentation | Updated November 14, 2026
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 4,137 automated checks before it ships, and every screen change is walked through before release.
What the Step Does
Ask My AI is an AI step you write the question for. You write a prompt in your own words, give it named inputs from earlier steps, and the step asks your own Claude account and hands the answer to the next step. It uses the Claude API key saved in Settings, AI Agents, and the model chosen there (AI Settings). MountDev Ops runs no model of its own, so nothing about it is billed by us.
For example, a prompt of "Write one friendly paragraph for [[site]] using these notes: [[notes]]." with the inputs site and notes.
The Prompt and the Inputs
- The prompt is up to 4,000 characters. Write
[[name]]where an input belongs. - There are up to 8 inputs. Each has a short name (lower case letters, digits, and _) and a value picked from an earlier step or typed. Up to 20,000 characters of each value are sent.
- A prompt that uses
[[name]]for an input that does not exist is refused when the workflow is saved. - An optional list of answers to give back (1 to 10, each a short name and a label) makes the step answer with those details, which later steps use by name. Without it the answer is one block of text, up to 16,000 characters.
What Stays Safe
- The prompt is your own words, written when the workflow is saved. It is never filled in from a run, so nothing a run reads can become an instruction.
- The inputs are handed over only as material to work on. Words in them that tell the AI what to do are not followed, and text that looks like the end of the material cannot close it.
- The step cannot send, save, or change anything. Whatever it feeds goes through the same checks, cards, and permissions as any step. An email to a client still stops for a person to confirm it.
- The answer is kept with the step so the next steps can use it, and is never shown in Run History. The key, the inputs, and the answer are never logged.
What It Counts and When It Fails
- Each try counts 1 toward Background work, retries included, like the other AI steps.
- With no key, the step stops at once and says where to add it.
- If Claude is busy, the step is tried again after the usual waits. A key Claude does not accept is not tried again.
- If every input is empty, nothing is sent and the step says there was nothing for the AI to work on.
- Test Run says the step would ask your AI and sends nothing. Test With Sample Data runs it for real on a sample.
Who Can Use It
People who may manage workflows. An AI agent connected to your account can place the step in a workflow it saves (workflows_save); the workflow is saved turned off, and a person turns it on.
Where to Find It
The step works now through an AI agent and the API. The Add a Step pop up does not offer it yet; its screen is Planned and waits for its design to be approved. A workflow that has one shows it as Ask My AI.
Draft Queue Planned
A Draft Queue is a module you pick where a workflow's drafts wait for a person to approve or discard them on the record. It builds on workflow steps that create records, which are not built yet.
