Ask the AI assistant

This page shows you how to use the AI assistant inside Deployment Manager. The assistant answers your questions while you work. You do not need to leave the screen you are on.

What the assistant is

This section says what the assistant does.

The assistant is a chat panel inside the tool. You type a question. It writes an answer.

You can ask it things like these:

  • What does this variable mean? What happens if I set it wrong?

  • Which services does this topology need?

  • Why did the check fail?

  • Why did the deploy fail, and what do I do next?

  • Where is the + New Installation button?

It answers from two sources:

  • our documentation: the deployment guides, the service contracts and the pages of this tool.

  • what it can see about the tool right now. For example, which installations exist. See What it can see, and what it never sees.

The answer can be wrong. Check it against the log or the screen before you act on it.

For the product words on this page, see the glossary.

Before you start

This section says who can use the assistant, and when it is not there.

  • You need the role user or higher. A viewer does not see the assistant. See the role user.

  • An administrator must turn the assistant on. If it is not turned on, the assistant does not appear on the screen at all. See assistant settings.

  • If the assistant service does not answer when the page opens, the button is also hidden.

If your role changes while you work, the panel closes and shows this message: "You can no longer ask the assistant here. Your role has changed, or your session has — sign in again, and ask an administrator for access to the tool if it stays this way."

Steps: ask a question

This section shows how to open the assistant and ask one question.

  1. Find the round button in the bottom right corner of the screen. Its tooltip is Ask the FT assistant.

  2. Press it. The FT Assistant panel opens.

  3. Click the text box. It says "Ask about a variable, a topology, a failing step…​".

  4. Type your question.

  5. Press Enter, or press the Ask button. To start a new line without sending, press Shift+Enter.

  6. Wait. Three dots show that the assistant is thinking.

  7. Read the answer. The words appear while the assistant writes them.

The first time, the panel shows four example questions. Press one to put it in the text box. You can change it before you send it. Or write your own question.

You can also open the assistant from a small ? button next to some fields. You find it next to variables, next to server variables, and on step 2 of the topology builder. Its tooltip says Ask the assistant about and then the name of the field. Press it. The panel opens and sends a question about that field for you.

What you see

This section describes the parts of the panel and of an answer.

The top of the panel has three buttons:

  • Export this conversation saves the whole chat as a file. The file name is assistant- and the date, for example assistant-2026-09-21.md.

  • Clear the conversation removes the chat and starts a new one.

  • Close closes the panel. The Esc key also closes it.

Under an answer you can see these:

  • Ask next: and up to three questions. Press one to ask it.

  • Show me: and up to six names of things on the screen. Press one. The tool highlights that thing on the page.

  • A copy button. It copies the answer. Some answers contain a command. That command has its own copy button.

  • Report a bug. See File a bug from an answer.

The chat stays when you open another screen of the tool. It also stays when you reload the page. The panel keeps it in your browser.

You can ask only one question at a time. If you send a second question too early, you see: "The assistant is still answering. Wait for it to finish, or clear the conversation."

Ask about a failure

This section lists the buttons that send a failure to the assistant for you.

The first two buttons appear only after a failure. They send the question and the log, so you do not need to copy anything. The other buttons are always there and send no log. Every button appears only if the assistant is available.

Screen Button What it sends

Deploy panel, under What to try next

Ask the assistant

The last 80 lines of the deploy log and the reason the panel found. If the deploy reached a server, the tool first collects diagnostics from the servers. The button shows Collecting… while it does this. The line under the buttons says: "Collecting diagnostics opens an SSH session to the servers; the assistant is also sent this log."

Test install, under What to try next or What to do next

Explain this failure or Explain with FT Assistant. Both do the same thing.

The end of the terminal output from the test install. If the test servers still run, the tool first collects diagnostics (about 20 s). If the servers are gone, it sends only the terminal output.

Training, on a step

Ask

A question about that training step. It puts the question in the text box. It does not send it. You press Ask in the panel.

Operations, People, the ? next to a figure

Ask the assistant about it

A question about that figure, and the definition of the figure.

Sometimes the assistant stops working after the button appears. Then, when you press it, you see: "The assistant is not available on this installation. The output above is the whole log." On Test install, the second sentence says: "The transcript above is the whole evidence."

For what these screens show, see Step 6. Deploy and Step 5. Test the bundle.

Attach a log yourself

This section shows how to give the assistant a log file from a server.

  1. Press the paper clip button in the panel. Its tooltip is Attach a log file. You can also drag the file onto the panel.

  2. Choose a text file, for example the output of collect-diagnostics.sh.

  3. A small label with the file name appears above the text box.

  4. Type your question and send it. The file goes with the question.

To remove the file before you send, press the x on the label. Its name is Detach context.

Only one file or log goes with a question. A failure button uses the same place. So a new file replaces the old one.

The tool checks the file before it sends it:

  • A file over 4 MB is refused. You see: "That file is N MB. Attach the text log itself (up to 4 MB), not an archive or a core dump."

  • A file that is not text is refused. You see: "That does not look like a text log. Attach the output of collect-diagnostics.sh, or the log itself."

  • A file the browser cannot open gives: "That file could not be read."

  • Values that look like passwords are replaced with *** before the file leaves your browser. The panel tells you how many, for example: "Attached file. 2 value(s) that read as credentials were replaced with *** before anything travels." The label also shows the number with the word masked.

A diagnostics file is long. The tool does not send all of it. It picks the parts that show errors.

What it can see, and what it never sees

This section says what information goes to the assistant with your question.

With each question, the assistant can see:

  • which installations exist, their customers, hosts and services.

  • how the installation in your question is built.

  • the values that people entered for that installation’s variables.

  • the log or file you attached, or the one a failure button attached.

It never sees a secret. Secrets are not stored in the place it reads from. A value whose name looks like a password or a key is also held back, whatever it contains. A very long value, for example a certificate, is always shortened.

An administrator decides two things:

  • whether installation names, customer names and hostnames are sent at all.

  • whether the values of variables are sent.

If they are turned off, the assistant can still explain the product. It cannot answer about your installation. See assistant settings.

The assistant can also run a check on a server for you. That check only reads. It works for a short time and only on one installation. See Step 7. Check that it works.

File a bug from an answer

This section shows how to send a bug report to Jira from an answer.

Every answer has a Report a bug button. The tool files nothing until you check the report and confirm it.

  1. Under the answer, press Report a bug. The button shows Writing…​ for a moment.

  2. A draft appears. It has a Title and a Report, built from the conversation so far. A log goes with it only if it was sent with this answer.

  3. Read the draft. Change the title or the text if you need to.

  4. Fill in every field the Jira project asks for.

  5. Check the line Attached:. It names the log file that goes with the ticket.

  6. If you see Still missing, add it yourself:, add that information to the report.

  7. Press Create in Jira. The button shows Filing…​. To stop, press Cancel.

  8. When it is done, you see Filed as and the ticket number. Press Open ticket in Jira to open it.

Things that can go wrong:

  • "Set jira.url, jira.token and jira.projectKey in Settings before reporting a bug from here." Jira is not set up. Ask an administrator.

  • "This Jira project requires fields Deployment Manager cannot know: …​" The missing fields appear in the draft. Fill them in and press Create in Jira again.

  • Any other error from Jira. Press Copy the report. Then paste it into Jira by hand.

  • "Could not reach the bug tracker." Try again later.

For what else a ticket needs, see what to attach to a ticket.

Limits and messages you may see

This section lists the messages the panel shows when it cannot answer.

When a question fails, the panel shows the reason and a Try again button. Try again asks the same question with the same log.

Message What it means

"Too many answers in flight right now. Try again in a few seconds."

Too many answers are being written at the same time. Wait a few seconds. Press Try again. The tool keeps some capacity free for deploys and Test installs.

"The assistant is unreachable right now."

The panel could not reach the assistant. Try again later.

"The assistant service did not answer. Try again later."

The tool could not reach the assistant service. Try again later.

"The assistant could not answer (HTTP number)."

The assistant refused without a reason. Try again. If it happens again, tell an administrator.

"The assistant finished without an answer. Ask again, in other words."

Ask the same thing with different words.

"The assistant closed the connection before answering. Ask again - if it keeps happening, the assistant service is restarting or overloaded."

Ask again. If it happens again, wait a few minutes.

"That answer stopped early - the assistant closed the connection before finishing it. Ask again if it looks incomplete."

The answer is not complete. Ask again.

Size limits:

  • Everything in one question shares one size limit. This includes your words, the attached log, and what the tool adds about its current state.

  • Only the first 4000 characters of your typed question are used.

  • A long log is cut to fit. The tool keeps the end of the log, because the error is usually at the end.

  • If a question is still over the limit, the assistant refuses it. It does not shorten it for you. Ask a shorter question, or attach a smaller log.

Good tips

This section gives tips for better answers.

  • Ask one thing at a time. You can ask the next question after the answer.

  • Name the installation. Write its name in the question, for example "Why did the deploy of qa11m fail?". Then the assistant knows which installation you mean.

  • Paste the exact message. Copy the sentence from the panel, word for word. Do not describe it in your own words.

  • Use the failure buttons. They send the log for you. See Ask about a failure.

  • Attach the log from the server. An answer is better when the assistant can see the real error line. See Attach a log yourself.

  • Check the answer. Compare it with the log or the diagnostics file before you change anything.

The AI Assistant tab

This section describes the admin view of the assistant.

The AI Assistant tab shows what people asked the assistant. It also shows which questions it could not answer. Only a global_admin sees this tab. The chats are removed after a month. See the in-app assistant and Operations journal.

If something goes wrong

This section says where to look when the assistant cannot help.