The message box
Attachments, the model picker, thinking mode, mentioning an agent and running a skill.
The box at the bottom says "What's on your mind?". Most of what you need is in it or one click away.
Typing and sending
Enter sends. Shift+Enter makes a new line without sending, which is what you want for a long brief.
Long messages are good. There is no prize for brevity here and a three sentence brief beats a five word one almost every time.
Attaching files
Click the paperclip, or just drag files onto the window, or paste an image straight from your clipboard. Once attached, each file shows as a small card above the text box with its name and a quick summary (page count for a PDF, sheet count for a spreadsheet, line count for a text file). Click a card to preview it, or the cross to remove it.
Per message you can attach up to 5 documents and 10 images.
If your spreadsheet has several sheets, Asteria asks which to include before uploading. Full detail on formats and limits is in Files in a conversation.
If the paperclip is greyed out and says your storage is full, you have hit your personal storage limit. Delete some old uploads, or ask your administrator for more room. See Usage.
Talking to an agent: the @ menu
Type @ (at the start of a message or after a space) and a list of your agents appears. Pick one and your message goes to that agent instead of to Asteria, with all of its instructions, collections and tools.
A banner then reads "Talking to name" and stays until you click Exit agent. See Agents.
Running a skill: the / menu
Type / and pick a skill to bring a saved set of instructions into the conversation. Use it for a repeated task ("rewrite this in our tone of voice") you would rather not explain every time.
The skill arrives as a pill in the message box, and it stays in force for the whole conversation, not just the message you send it with. To stop using it, delete the pill. See Skills.
The thinking button
The brain icon appears only on models that let you choose. Turn it on and the model reasons at more length before answering. It is slower and costs your organisation more, so leave it off by default and turn it on for genuinely hard problems: an argument with several steps, a tricky calculation, an ambiguous document.
For "summarise this" or "make a chart" it changes nothing worth waiting for. If you do not see the button at all, the model you are on has already decided for you.
Choosing a model
The picker shows the models your organisation has made available, by their display names. You can leave it alone. Your administrator set a default that is right for almost everything, and Asteria will pick suitable models for background work regardless of what you choose here.
Depending on how your organisation is set up, models may carry a small label: Fast, Balanced or Most capable. That is the whole vocabulary you need. The labels say what each one is for, so you never have to know what a given model name means.
Reasons to actually change it:
- You attached an image and it says the model does not support images. Switch to one that does.
- Answers feel shallow on a genuinely hard task. Try a bigger model, or one labelled for intelligence.
- You are doing something long and repetitive and speed matters more than polish. Try a smaller one, or one labelled fast.
That is the whole list. Model shopping is not where the quality is: a better brief on the default model beats a bigger model on a vague one, every time.
If you see a warning above the message box about a fallback model, nothing is wrong. The model you picked was briefly unavailable, so your organisation's backup answered instead, and the notice names both so you are not left wondering why the reply reads differently. Your answer is above it as usual, and your choice in the picker is unchanged. Dismiss it with the cross.
Worth mentioning to your administrator if you see it often: a model failing repeatedly is something they can fix, and they may not have noticed yet.
Send and stop
The button on the right sends. While Asteria is working it becomes a stop button, and stopping is a perfectly normal thing to do: if you can see it has misunderstood twenty seconds in, stop it rather than waiting for a wrong answer.
You can also just keep typing while it works. Send another message and it is picked up as a course correction on the run in progress rather than queued for later. Send a great many in a row and it will ask you to wait for the current answer to finish.
When it asks permission
Some actions need your say so before Asteria takes them. The conversation shows an Approval needed card with Allow and Deny (Enter and Esc also work). The message box is disabled until you answer.
If you ignore it, it is declined automatically after a few minutes. Declining is safe: Asteria carries on without that step and tells you what it could not do.
Keyboard shortcuts
Defaults, all changeable in your profile:
| Shortcut | Does |
|---|---|
| ⌘K / Ctrl+K | Search your conversations |
| Shift+? | Show this list in the app |
| N | New chat |
| / | Jump into the message box |
| Esc | Leave the message box |
| A | Attach a file |
| ↑ | Edit your last message |
| C | Copy the last answer |
| R | Regenerate the last answer |
The single letter shortcuts only fire when you are not typing in the box, which is why Esc then C is the fast way to grab an answer.