Talking to it

When it answers#

Xertron does not read every message in your server. It answers in three situations, and stays quiet otherwise, so a busy channel does not turn into a conversation with a bot nobody invited.

• You mention it. Put @Xertron anywhere in the message, in any channel it can see.
• You reply to it. Reply to one of its messages and it picks the conversation back up, with no mention needed.
• You are in one of its threads. Every message in a thread it started is meant for it.

When it starts working it posts a short status message, then edits the finished answer into that same message. That is why it needs the Bypass Slowmode permission: in a channel with slowmode on and that permission missing, the status line is the only message it ever gets to post.

It also ignores other bots and webhook apps by default, on every plan. Their messages are invisible to it, and they can never trigger a reply. An admin can let it read specific apps from the dashboard, but even then, only a person can make it respond.

Long answers

Xertron always replies in one message. It never splits an answer across several. When an answer is too long for a Discord message, it posts a fresh message carrying a single response.md file instead. Discord previews the beginning of that file right in the channel and expands the rest when you click it, so there is nothing to download to read it. The reasoning and the usual model footer go inside the file. It remembers the whole answer either way, so follow-up questions about any part of it work normally.

Threads#

A thread is the right place for anything longer than a question and an answer. Run /startthread with your opening message, or right-click any message and choose Start a Thread to open one from it.

A thread keeps its own history back to the moment it was created, and it can carry its own persona, model and thinking depth for the whole conversation. Every message in it reaches Xertron without a mention.

Xertron puts a 10 second slowmode on any thread it starts. That is not a punishment. Two messages arriving while it is still answering the first one produce a worse answer to both, and the gap is enough to stop them overlapping.

In a regular channel there is no thread history to draw on, so it reads a rolling window of up to about 100 recent messages, capped by a character budget rather than a fixed count.

Steering one reply#

Flags are typed inline, right after the mention, or at the start of your message in a thread. They apply to that one message and change nothing else.

Every reply already reasons at least a little, so there is no flag for less than !think-low and no way to turn reasoning off. What !think adds is a Show reasoning button on the reply.

You can use one model flag and one thinking flag per message. Two of either conflict, and Xertron says so and waits rather than guessing which one you meant.

Each reply carries a small footer saying what was actually used, like -# model: flash. The thinking part only appears when it was above the standard level, so an ordinary reply's footer stays short.

To change how your messages behave by default instead of one at a time, run /settings. See Commands and flags.

Starting over#

In a regular channel, send @Xertron !clear and it forgets the conversation up to that point. It reacts with a broom to confirm, and starts fresh from your next message.

The mention is what makes it a command. A bare !clear on its own does nothing, so a message about clearing something never wipes a conversation by accident. Replying !clear to one of its messages works too, since that also addresses it.

Inside a thread there is nothing to clear, and Xertron will say so. A thread remembers everything back to its creation by design. Leave it and start another one instead.

Files and images#

Attach a file to any message that reaches Xertron and it reads it. Text files, PDFs and Word documents all work on any plan. Images need Premium, because reading a picture is a different and more expensive kind of model call.

Past about 5,000 characters it stops reading a file front to back and starts searching it by meaning, so asking for the part about the refund policy finds it whether or not you remember the wording. That works across every large file shared in the server, not just the one you are looking at, and it is part of the same file tool on every plan.

Sending a file with nothing else

Attach a text file, PDF or Word document with no message text at all and Xertron stores it quietly. No reply, no reaction, and nothing drawn from your server's usage. Ask about it in your next message and it already knows what is in it.

An image sent the same way still gets a full reply, since describing it is the only useful thing to do with a picture nobody asked a question about.

If storing a file fails, Xertron says so rather than forgetting it silently. Send it again.

When it holds back#

Xertron steps down in capability as a server approaches its monthly limit, rather than stopping dead at it. If a reply feels lighter than usual, that is usually why, and it will say so plainly when asked.

Send messages faster than it accepts them and it asks you to slow down instead of answering. Wait a moment and try again.

An admin can also turn on eco mode, which quietly holds everyone to the fast model with no extended thinking to conserve the monthly limit. Quick, simple replies on a server with eco mode on are the setting working, not a fault.

What your own messages are allowed to reach also depends on your roles. See Roles and limits.