Your first chat

Context Code revolves around one simple idea: you open a folder on your disk (your project) and talk to an agent that can actually work on it — read files, edit them, run commands. This page takes you from startup to your first answer.

The welcome guide does it for you

The first time you open the application, a five-step guide appears that gets it ready to work. Every step can be skipped, and you can see it again whenever you want from your account menu (“Show the guide again”) or in Settings → General.

1. What language shall we speak? — the app’s 17 languages. You can change it later.

2. How do you want to use AI? — three paths:

   • “I already have a Claude, ChatGPT or Copilot account” — your browser opens, you sign in with your usual account and the app detects on its own that you’re done. If the provider asks for a device code, it shows it on screen.
   • “I have an API key” — six providers with a free tier to get started, each with its “Where do I get the key?” link. The Test button makes a real request and answers “It works: N models available.” or the specific error.
   • “I want an AI on my computer, no accounts” — it checks what your machine can handle, recommends a local model and downloads it with a progress bar. If your machine falls short, it tells you and suggests the other paths.

3. What are you going to work on? — you pick a folder (and the agent will be able to read and write there) or carry on without one with a plain chat.

4. Want to talk to it and have it read to you? — test the microphone for three seconds and let it hear how it sounds with the reading voice.

5. That’s it — it suggests three first questions you can click on.

When it finishes, the chat is already set up with the profile you just connected.

The rest, step by step

1. Open (or create) a project.

   From the sidebar, pick the folder you want to work with. It can be an existing code project or an empty folder where the agent starts from scratch. Your recent projects stay in the sidebar so you can come back with one click.

   

   The sidebar: your projects and chats, always within reach.

2. Connect an AI provider.

   In the chat’s engine selector (or in Settings) you will see the list of Available Engines & AI, each with its status: Connected or Unauthenticated. Connect whichever you prefer — with Anthropic (Claude), OpenAI or GitHub Copilot you just sign in through the browser (OAuth); Gemini, xKiro and the rest use an API key; and the local ones (Ollama, LM Studio, Context Forge) need no account at all. The full breakdown is in AI providers.

3. Pick the model.

   Next to the chat’s input field is the provider and model selector. You can change it at any time, even mid-conversation: a cheap model for the easy stuff, a powerful one for the hard parts.

4. Pick the approval mode.

   The permissions selector sits in the same chat composer: Read only, Ask for approval, Approve for me or Full access. It defines what the agent can do without asking you. If this is your first time, Ask for approval is a good starting point. Full explanation in Approval modes.

5. Write your request.

   Ask for what you need, in your own language and with context: “explain what this project does”, “add validation to the signup form”, “find out why login is failing”. The agent shows every step in the chat: which files it reads, which commands it runs and what changed.

   

   The agent shows every tool it uses: reads, edits, commands.

Tricks worth knowing from day one

• Attach files and images to your message (or drag them into the chat) to give the agent context.
• Voice dictation: you can talk to it instead of typing — see Voice. And if you want a back-and-forth conversation, with interruptions and a spoken answer, turn on Real-time conversation.
• Ctrl+Shift+F searches text across the whole project; the quick file and command finder also opens panels and actions — see Search.
• Stop: if the agent is heading the wrong way, the stop button cuts the current turn short.
• The token usage panel shows how much you have consumed per project and per model — see Token usage.

One tab per chat

Chats open as tabs in the same bar as the editor files and the SQL console. The tab is called “Project <name> - <chat>”, or “Chat - <name>” when the chat has no project; the width is fixed, so a long title is trimmed and you see it whole on hover.

• A click in the sidebar opens that chat’s tab, or jumps to it if it was already open.
• Activating a tab switches to that chat’s project: file explorer, editor and Git panel come along.
• A green dot (“Working”) means the turn is still running; a ”!” (“Waiting for your answer”) means the agent has asked you something.
• Ctrl+Tab and Ctrl+Shift+Tab walk the whole bar, chats and files alike; drag them to reorder.
• Closing the tab does not stop the chat: not with the ×, not with a middle click, not with Ctrl+W. If it was working you get a warning —“The chat keeps working”, “Closing the tab stops nothing: the chat keeps working in the background and you can come back to it from the sidebar.”— with the “Don’t warn me again” checkbox.
• Right-clicking a tab offers “Close the tab (the chat keeps working)” and “Close all chat tabs”.

The preview tab

Browsing chats with one click used to fill the tab bar. That is why there is a single provisional tab, like in a code editor: a single click on a chat in the sidebar opens it there, with the title in italics (”… (preview: double-click to keep it)”), and the next single click on another chat reuses that same slot, without moving it. There are never two provisional tabs.

The tab is kept by any of three gestures: double-clicking the chat in the sidebar or the tab itself, typing a message in that chat —it is no longer a peek— or dragging its tab. Going back to an already-kept chat does not demote it.

You turn it off in Settings → General with “Open chats as a preview with one click” (on by default, and disabled if “Chats as tabs” is off). Turned off, every click opens a permanent tab; and turning it off keeps whatever provisional tab was open.

Open tabs are remembered and come back when you reopen the app. If you prefer the old behaviour —a single chat at a time—, turn off “Chats as tabs” in Settings → General; turning it off closes the open chat tabs.

Pinning and archiving chats

In the sidebar, right-clicking a chat —including those on the “Chat” tab, the ones with no project— opens a menu with “Pin chat” / “Unpin chat” and “Archive chat”.

• Pinned chats move to the top of the list, the last one you pinned first, and stay there after a restart (it is stored in the database, not in the browser).
• Archived chats leave the working list without losing their history: they wait in the “Archived” section, with their count beside it, and come back with “Bring back to the list” or “Restore all”. If you archive one that was still working, it is marked “Archived, but still working”.
• The sidebar search finds archived chats too: the result carries a “Restore and open” button.

The status icon in the top bar

In the top bar there’s a heart that sums up whether the application is ready: “All set”, “It works, but something needs a look” or “Something is missing before it can answer”, with the number of warnings.

It checks four things: that there’s a connected AI with credentials, that the chosen model still exists in your account, that there’s internet and that the helper programs are in place (the dictation one, for example). Every warning comes with a Fix it button that takes you straight to the missing step of the guide.

Tip

If the chat “does not respond”, it almost always means the active provider has no valid credentials. Open the status icon: it tells you in one sentence and fixes it in one click. More cases in Troubleshooting.

More of this screen

A brand-new chat suggests where to start: code, SQL, Markdown, graph or a team of agents.

When you open a new project, Context Code shows what it can do before the first question.