While maxinja works

Your brief becomes saved work as maxinja builds it. Follow the progress, steer a decision while it works, and inspect the resulting campaign or journey. This page explains how changes, interruptions and follow-ups affect that work.

It changes the real thing, not a draft

Two habits are worth building:

• Say up front how far you want it to go. "Set it up but leave it as a draft, I will send it myself" is a normal thing to ask for.
• Open the thing itself afterwards and check what is expensive to get wrong: the audience, the schedule, and the sender.

Following what it is doing

maxinja hands each piece of work to a specialist that knows that one thing (an email, a broadcast, a workflow) and reports back. You see named jobs open and close with a one-line result, progress lines while a job runs, and the reply typing itself out.

What it reports is what changed, in words: which schema the audience filters, the condition in plain English, how many contacts match. The query behind it stays on the server, and neither you nor the API ever has to look at it.

Answering its questions

maxinja can ask for missing information or a decision during a task. The question appears in the conversation with its own input, and that is where your answer goes: the normal message box waits until you have replied.

• Some questions come with answers you click instead of typing.
• Some come with a countdown, and pick a stated default if you do not answer in time.
• Some ask for a file, or for you to point at something with @.

One request at a time

One exchange runs at a time. A message sent while maxinja is working becomes steering input for that active work. State the correction and the part it changes, such as "keep the offer fixed; only adapt the introduction." You can also stop the exchange.

Stopping leaves completed changes in place. Open the saved work before continuing, then ask for the specific correction, such as "put the delay back to 3 days." Stopping or retrying does not reverse a send or a publication.

To correct the request itself rather than the result, edit an earlier message and retry it. That deletes everything that came after it in the conversation and runs again from there. It rewrites the conversation, not your space: whatever maxinja already built is still there.

Steer, stop and retry with the saved state in mind

A message sent while work is active is queued as steering and consumed as the running assistant reaches a checkpoint. It is not an immediate undo of a tool call already in progress. If a question is pending, a submitted answer resolves that question instead.

Stop requests interruption and clears queued follow-up input. It cannot recall an accepted email, undo an external tool effect or roll back a completed edit. If the connection drops, reopen the conversation and inspect its saved status before sending the same brief again.

Editing and retrying an earlier message removes later conversation exchanges and starts a new exchange with the revised request. Deleting messages only changes the conversation. Neither action restores earlier versions of campaigns, records or workflows. Refer to the existing saved item when continuing so that a retry does not accidentally become a request for a duplicate.

If maxinja reports a transcript limit, start a new chat and attach the existing item. If it reports a full input queue, wait for the queued input to be consumed before adding another correction.

Follow-ups keep working on the same thing

Once maxinja has built you a broadcast, the next request normally refines that same broadcast rather than starting a new one. When you do want a second one, say so: "now make a separate broadcast for the trial users".

Ask it questions about what it built, too. "Why does this workflow have two branches?" goes to the same specialist and comes back as an explanation, without changing anything.

Emails it writes for you

When the work is an email, maxinja writes the content, keeps a preview rendered from sample data, and looks at that preview to check its own work. It follows the selected brand kit for colors, fonts, and images.

If the email needs a picture your brand kit does not have, maxinja has two ways to fill the slot, and it chooses between them itself:

• A labeled placeholder. A stand-in that blocks out the slot at no cost. maxinja reaches for this while the layout is still moving, or when the picture is yours to supply.
• A generated image. A real image from the image model, on brand. This one costs credits.

To decide it yourself, say so in the request ("use placeholders, I will add the images"). To replace either afterwards, add the picture you want to the kit's asset collection, and maxinja uses that instead of inventing one.

Check what completion means

maxinja checks the saved result before reporting a completed change. A completed draft edit can still have unrelated draft issues; it does not mean the campaign was sent or the workflow was published. Confirm those states in the campaign or workflow itself, and use email or run history to inspect execution.

When something goes wrong

• A request fails. It is marked as failed with the reason. Nothing is rolled back: whatever was already built is still there, and a follow-up message can continue from it.
• Credits run out. maxinja checks your credit balance before it starts and again as it works, so an exhausted balance stops a run partway rather than at a tidy point. Top up, then ask it to carry on.
• You do not have write access. Workspace assistant commands and execution are refused. If access is removed during work, further execution is stopped; previously saved work remains.

Related

• maxinja: what it can and cannot build.
• Working with the assistant: the chat pane, references, and attachments.
• Broadcasts and templates: the things it builds most often.
• Adaptyle: the per-recipient personalization it writes into an email.