Appearance

Start a chat ​

A chat is one conversation with one agent. Here's how to begin.

  1. Open a new chat from the sidebar. EchoMode opens a short setup wizard so you can pick the agent you want to talk to — see Choosing an agent.
  2. Type your message in the composer at the bottom of the chat. The field reads "Type your message…" until you start typing.
  3. Attach files (optional). Select the + button next to the composer to attach a file. EchoMode accepts images, PDFs, Word documents (.doc/.docx), and text files (.txt, .md, .csv). Attached text shares a slice of the model's context window, so a very long file is trimmed and the assistant is told it was — put it in a Dataroom instead when you want the whole document available.
  4. Send. Select the send button (the up-arrow) or press Enter. The button is unavailable while a reply is still streaming or while the field is empty.

The chat composer

The composer: type your message, attach files with the + button, and send with the up-arrow.

What happens after you send ​

The reply streams in word by word. When it finishes, EchoMode scores how on-topic the exchange was — you'll briefly see "Computing alignment score…" in the header while it works. That score powers the drift signals described in Understanding Drift.

Long conversations ​

Every model has a limit on how much text it can read at once. The bar above the composer shows where you are against it.

When a conversation approaches that limit, EchoMode summarises the older turns and sends the summary in their place, so the chat keeps working instead of stopping. Your transcript is not changed — scroll up and every message is still there, word for word. Only what the model re-reads each turn is condensed, and the bar shows compacted once this has happened.

Very occasionally a single message cannot be made to fit — usually a large attachment plus a wide Dataroom selection. EchoMode will say so and ask you to remove a file or narrow the selection, rather than sending something the model would reject.

TIP

Want the assistant to answer from your own documents? Attach them through the + menu's Dataroom picker — see How search works.