> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mossx.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Chat area
> Read streaming replies, thinking, tool calls, the context ledger, and the status panel.
The chat area is the main surface of the desktop app. Use it to see how the assistant thinks, which tools it called, which files it changed, and how much context this turn used.
## Empty session
A new session with no messages shows a welcome / empty state that asks you to send a message to the current engine.
After you configure a provider and select a workspace, the input is ready.
## What a message contains
Assistant messages are usually several blocks:
| Block | Meaning |
| ---------- | ----------------------------------------------------------------------------------------------------------------- |
| Thinking | The model's internal reasoning. You can collapse it. Long thinking uses a live channel so the UI stays responsive |
| Body | Markdown reply. You can copy code blocks |
| Tool calls | Read files, edit files, run commands, search, browse the web |
| Diff | File change comparison |
| Usage | Token / cache / estimated cost when the runtime provides metadata |
Long sessions render only a recent DOM window by default. Scroll up to page in older messages so the chat does not get slower over time.
## Streaming and stop
Streaming is on by default. While the model generates, the send button becomes **Stop**.
If you stop it yourself, the completion toast and sound do not play.
## Rewind and fork
When the runtime supports it, Claude / Codex provide:
* **Rewind**: go back to before a message
* **Fork**: start another conversation line from that point
Rewind can affect later messages and files. Review the list before you confirm.
## Context usage and ledger
Context usage appears near the input. The app also has a **context ledger**:
* Selected or inherited context sources
* Token / character estimates
* Freshness and attribution confidence
When you are close to the limit, you can:
* Send a compact / summarize command (depends on the runtime)
* Start a new session and attach only the files you need
* Use `@` to attach a file instead of pasting a long log
## Status panel
Longer tasks show a status panel:
| Content | Notes |
| ------------- | --------------------------------------- |
| Todo / Plan | Todos or plan steps the model split out |
| Subagents | Parallel subtasks |
| Files / edits | Files changed in this turn |
## Permissions and questions
When the model needs to edit a file, run a command, or ask you a multiple-choice question, a dialog appears. If a path is outside the allowed directory, Claude may also show a Directory Grant card.
Details: [Modes and permissions](/en/desktop/chat/modes).
## Related pages
`@`, `/`, `$`, `#`, `!`, `@@`, `@#`.
Sidebar sessions, history, and rewind.