A prompt is simply a text task for the agent, such as “gather what you can find about company N”. The prompt queue lets you sketch out a list of such tasks in advance, and the agent carries them out one after another by itself — no manual sending of each one. Handy when the same kind of process needs to run many times: prepare a dozen rows, go about your business, and come back to finished results.
What it is and why
The queue is just a list of rows. Each task runs in its own chat, and the chat’s title is taken from the task text. Type a row, and you can send it manually right away — or leave it in the queue and let AutoSend handle it.
There is one queue for everyone: both Research Agent and the Xedant Agent interface use it. These are the same rows in one shared arrival order — add a task in either interface and it is instantly visible in both.
Where to find it
The queue has two entrances, and both open the same dialog:
- the prompts button in the right part of the chat’s message input;
- the “Prompts (N)” button at the bottom of the documents sidebar — the counter shows the current number of rows in the queue.
Working with rows
The queue dialog is a list of rows, and every row is fully under your control:
- adding — the “New Prompt” button creates an empty row; the first row can simply be typed into right away, with no extra clicks;
- editing — the task text, model and skill are changed right in the row;
- order — rows move up and down; new rows are added to the top of the list, while sending goes from the bottom up, so a row added earlier goes first. Move a row lower, and it will go sooner;
- deleting — an unneeded row can be removed at any moment.
Changes save themselves — there is no “Save” button: half a second after you stop typing. New rows pick up the model and skill currently selected in the chat by default.
The prompt generator
When there are many tasks and they all follow the same shape, you do not need to type each one by hand: the “Generate Prompts” button turns a pattern and a list of values (one per line) into ready-made queue rows.
Substitution follows one simple rule — the place for a value in the pattern is marked with the word NAME:
- the pattern contains NAME — the value from the list is substituted in its place;
- the pattern has no NAME — the value is appended to the pattern with a space;
- the pattern is empty — the value is added as-is.
Line order is preserved: the first line of the list ends up at the bottom of the queue and runs first. For example, the pattern “Research the NAME market” with the values “solar panels” and “wind turbines” creates two tasks — and “Research the solar panels market” goes first.
Manual send
Every row has a send button: one click, and the task goes into a new chat. The chat title is taken from the first line of the task (up to 100 characters), the model and skill set in the row are applied, and the app opens the new chat right away.
The row is removed from the queue only after a successful launch. If something stopped the chat from being created, the row stays in place — just try again.
AutoSend
Turn on AutoSend, and the agent starts taking tasks from the queue on schedule, with no part left for you. The queue keeps working even when every tab is closed: leave, and come back to ready results. The same settings popover carries two more switches borrowed from the Agent — “Auto-commit before sending” (saves the project state in Git before each send) and “AutoFix disabled until done” (keeps Xedant Agent’s automatic build repair from firing in the middle of a running queue). The core settings next to the toggle:
- Max active chats — how many tasks can run at the same time, from 1 to 10 (default 3);
- Check every N seconds — how often the agent looks into the queue, from 1 to 3600 seconds (default 10);
- Auto redirect — whether to follow into the chat AutoSend has just launched (on by default).
One pleasant extra: the moment one task finishes, the next one starts immediately, without waiting for the next check.
Peak hours
The “Send during peak hours” setting (on by default) is about the load on the models. Switch it off, and tasks for models in their peak load hours are not sent automatically — they wait in the queue — while tasks for off-peak models keep going out immediately.
Sync
The queue lives in the agent’s settings, so everyone sees the same thing. A change from any tab, any device, or the Xedant Agent interface itself appears everywhere instantly — and back. While you are typing, other people’s edits never get in the way: they apply only after you finish typing. As a safety net, the list is additionally refreshed from scratch every five seconds — even if a change notification gets lost along the way.
Licensing
When the license is invalid, adding rows and sending tasks are blocked: the app opens the licensing dialog and explains that the change was not saved. Viewing the queue keeps working without an active license.