# Chat: getting work done with the agent

Chat is where all the work happens on Quodara. Every control on the composer, the message area and the header, explained.

It is also where documents, images, video and reports come out, so it is worth knowing what each button actually does.

## Starting a chat

The first sidebar item, **new chat**, starts one; you can also open [Chat](#/chats) directly. Hand the same message to a **scheduled task** and a chat gets started for you when the time comes (see `08-schedule.md`).

1. Pick the **app** you want to work on in the composer
2. Type what you need
3. Press **Ctrl + Enter**

Choosing the official **default** app gives you a general assistant instead: it answers questions about the platform and helps with documents, media, marketing content and comparing approaches. It is not bound to any repository.

## What the composer can do

- **`@` references** — typing `@` opens a picker for **an app**, **a file this chat produced**, or **a file you uploaded**. Referencing an app lets the agent read its repository and deployment config directly, which beats describing it from memory.
- **Attachments** — multi-select upload, and you can paste screenshots straight into the box. Uploads go to the agent; anything it produces comes back below the message.
- **Merge into the app's main version** — folds this chat's edits into the repository's default branch.
- **Stop** — interrupts the current run and drops your original message back into the composer so you can rephrase.

## What you see while it runs

Message status moves through `Queued` → `Running` → `Done`, or `Stopped` if you interrupt it.

Chat is **not token-by-token streaming**, but you are not left waiting either: while the agent works, the message area shows its **execution steps** in real time — the files it read, the commands it ran, the capabilities it invoked. Until the first step arrives, a waiting animation stands in.

## What can appear in a reply

Replies are more than text. You will regularly see:

- **Interactive cards** — when the agent needs you to choose, fill in a form or acknowledge a risk, it renders a card in the message; tick and submit rather than writing prose
- **Artifacts** — generated images, music, video, Word / Excel / PowerPoint / PDF files. Images get a large preview, everything else is listed by filename for download.
- **Interactive reports** — long-form analysis, research and review output is rendered as a report card with a table of contents, charts and sortable tables; it opens in a side panel and can be shared
- **Preview and Publish** — one row of buttons per app the message actually changed
- **Merge requests and deployment records** — open one to see the log and result of that deployment

How to ask for images, music and video — and how to revise them — is in `10-media.md`; Word, Excel, PowerPoint, PDF and interactive reports are in `11-documents.md`.

## The four header actions

| Button | What it does |
|---|---|
| Collect | Bookmarks the chat; find it again under the **Collect** tab in the sidebar |
| Share to Blog | Posts the conversation to Explore, readable without signing in |
| Share as image | Exports the whole conversation as one tall image |
| Support ticket | Opens a ticket carrying this conversation as context |

## Asking well

The same request phrased differently gets very different results. A few things that reliably help:

1. **State the goal, not the implementation.** "Let users filter orders by date" beats "add a date picker" — the agent picks an implementation that fits the codebase.
2. **One thing per message.** Five unrelated changes in one request make it hard to tell which step broke something.
3. **Bring evidence.** Paste the screenshot, paste the red text from the console. It is far more useful than "it doesn't work".
4. **Preview before you merge.** Preview is a free chance to catch it; don't skip straight to publish.
5. **Ask for a plan first on big changes.** "Don't write code yet — give me two options and a recommendation" is a good opening move.

## Common questions

**Q: Can I continue a chat after stopping it?**
A: Yes. Stopping only interrupts that run; the context and the branch are intact, so the next message picks up where it left off.

**Q: Can I run several chats on the same app at once?**
A: You can, but each works on its own temporary branch, so chats touching the same files will conflict at merge time. Stagger them.

**Q: Are uploaded files kept?**
A: Attachments and artifacts stay with the chat and can be downloaded any time. For media the app itself needs long term, ask the agent to put it in the app's object storage.

Read next: `04-publish.md`