The chat window

Everything you need on one screen, nothing you don’t. The window is a chat; the power is in what hangs off it.

A Companion conversation has four regions: the model picker at the top, the message stream in the middle, the composer at the bottom, and the sidebar for chats and projects.

The composer

Where you type. Beyond plain text:

• Attachments — drop in files and images (on vision-capable models).
• Voice — push-to-talk on Space, or full Talk mode for hands-free.
• Send / stop — Enter sends, the button becomes Stop mid-stream.

Message actions

Every message carries inline actions on hover:

• Edit a user message and re-run from that point — the rest of the thread is replaced.
• Regenerate an assistant message — same prompt, fresh answer.
• Copy the raw markdown; code blocks have their own copy button.

Editing or regenerating forks cleanly: the conversation rewinds to that turn, so you can explore alternatives without starting over.

The stats footer (StatsRow)

Under each answer, a thin row of real numbers:

TTFT 2.5s · Duration 12.1s · Prompt 25 tok · Completion 824 tok · Speed 68 tok/s · Cached 22 tok (88%) · Model: …

• TTFT — time to first token. Speed — decode tok/s. Cached — prefix cache hits (the cross-turn KV win).
• Model — the model that actually answered. When you route through CoeOS, it shows the employed model, e.g. CoeOS · python — MiniMax-M3.

Projects

The sidebar groups chats into Projects. A project shares:

• a system prompt applied to every chat in it,
• memory toggles (personal and per-project),
• optionally a project wiki injected as context.

Move a chat into a project and it inherits that context. Projects are how you keep a long-running piece of work coherent across many conversations.

Read next

• Chat basics — sending, editing, regenerating, attachments in detail.
• Model picker — choosing models, Easy / Advanced / Expert.
• Memory — what Némo remembers and what gets injected.
• Voice — push-to-talk and Talk mode.