Code User Guide

Code is Artificial Labs' cloud development platform. You describe what you want; an AI coding agent plans, writes, runs, and verifies the work on a real cloud VM while you watch — with a full IDE in your browser.

1. Getting started

Signing in

• Go to your panel URL and click Sign In.
• Authentication is handled by Artificial Labs SSO — there is no separate password.
• Your session persists between visits; use the profile menu to sign out.

Creating your first app

1. From My Apps, click New App.
2. Give it a name and a short description of what you want to build.
3. Pick a model (you can change it later) — this is the AI that will do the work.
4. The platform provisions a cloud VM automatically. When the status shows ready, click the app to open the IDE.

Each app = one project + one VM + one ongoing conversation with the agent. Apps are private to you by default.

2. The IDE

Opening an app gives you a four-panel workspace:

File Explorer (left)

• Browse the project on the VM. Create, rename, and delete files and folders.
• Search across files.
• View revision history for a file.
• Right-click a file for AI helpers: "Explain this file" and "Debug info".

Editor (center)

• Monaco editor (the same engine as VS Code): syntax highlighting, multiple tabs, automatic language detection.
• Diff view: when the agent edits files, you can review changes side-by-side.

Terminal (bottom)

• A real SSH terminal into your app's VM. Open multiple terminal tabs; sessions survive page refreshes.
• Live system metrics (CPU / memory / disk) for the VM.

Chat (right)

• Where you talk to the agent. Everything the agent does shows up here as it happens.

On phones and tablets the panels collapse into a swipeable, touch-friendly layout.

3. Working with the agent

Starting a task

Type what you want in the chat box and send it. Be as specific or as vague as you like — the agent will plan, break the work into tasks, and start executing. Good prompts include the goal and how you'll judge success ("add a login page; it should reject bad passwords and redirect to /dashboard on success").

What you'll see while it works

• Status line — what phase the agent is in.
• Thinking box — the agent's reasoning streamed live before it acts.
• Task list — the agent's own checklist, updated as items start and complete.
• Action cards — every concrete step (file edit, terminal command, service restart, browser check) appears as a card with its result. File edits include a diff.
• Iteration counter — agents work in iterations (plan → act → check results). Each app has an iteration limit per run (default 20).
• Token usage meter — live context usage.

Steering and stopping

• Steer mid-run: send a message while the agent is working — it's queued and injected as guidance at the next opportunity.
• Stop button: halts the run immediately. Progress made so far is kept, and the agent saves a snapshot so the next run can resume context.

Attachments

• Files: drag-and-drop or use the attach button to give the agent files as context.
• Images: paste or upload screenshots — vision-capable models will read them.

Reviewing changes

The pending-changes panel lists every file the agent modified in a run. Review the diff for each and keep or discard changes individually, or accept/reject in batch.

Conversation management

• The full conversation persists across visits and syncs live across browser tabs.
• Long conversations are summarized automatically.
• Clear chat wipes the conversation for a fresh start (the code on the VM is untouched).

Memory and continuity

• App memory: the agent accumulates durable facts about your project and recalls them in future sessions.
• Run snapshots: if a run ends early, the agent saves its task list and working state and picks it back up on the next run.

Skills and shortcuts

• Skills are reusable prompt playbooks for common jobs.
• Shortcuts are slash-style macros in the chat box that expand into full prompts.

Choosing a model

The model dropdown in the app header switches which LLM powers the agent. Different models trade off speed, cost, and capability; you can switch any time between runs.

4. Running and previewing your app

• Browser preview: the built-in browser tab renders your app running on the VM — with URL bar, back/forward/refresh, zoom, and developer tools (console output, network requests, storage).
• Ports & proxy: apps are reachable through an authenticated proxy; you can attach a custom proxy domain in app settings.
• Visibility & ACL: apps are private by default. You can make an app public or whitelist specific IPs.

5. Account, billing, and admin

Profile

• Edit your name and contact details. Email comes from SSO and is read-only.
• Toggle dark/light theme (remembered across sessions).

Billing

• Usage summary (VM hours, bandwidth, storage).
• Invoices with downloads, payment method and billing contact management, CSV exports.

6. vCode desktop integration

Code has a desktop companion, vCode, that runs the same agent against code on your machine instead of a cloud VM.

• Download vCode for Windows or Linux from the panel's vCode section.
• The panel shows all your vCode sessions: status, model, message and iteration counts, token usage, estimated cost, and last-active time.
• Sessions started on desktop are visible here, and cloud-side history survives even when your laptop is offline.

7. Tips for best results

• One goal per run. Small, verifiable asks finish faster and cleaner.
• Tell the agent how to verify. "…and confirm the endpoint returns 200" gives it a finish line.
• Steer early. Send a steering message rather than waiting for the run to finish.
• Use the diff review. Skim pending changes before building on top of them.
• Watch the token meter. Let auto-summarization do its thing, or clear chat when switching topics.
• Screenshots beat descriptions for UI bugs — paste the picture.

8. Troubleshooting

• Run ended without finishing — it likely hit the iteration limit. Its snapshot is saved; just say "continue" to resume, or raise the limit in app settings.
• Terminal disconnected — refresh the page; terminal sessions are persistent and will reattach.
• Preview not loading — confirm your app process is actually listening (check the terminal), then refresh the preview tab.
• Stream dropped mid-run — reconnects are automatic and replay anything you missed; the agent keeps working server-side.