Getting Started
Quick Start
Sign up, create a sandbox, run your first agent. Five minutes.
This walkthrough takes you from a fresh account to an agent running in a sandboxed workspace. The whole point is that you can hand it work and walk away. The platform keeps your agent running off your laptop, overnight if you want.
Step 1. Sign up
Visit console.ellul.ai/sign-up. Create your account with email or one of the supported OAuth providers. Pick a plan; you can change it later.
Step 2. Create a sandbox
Once signed in, you land on the dashboard. Click New Sandbox in the sidebar.
You will be asked for:
| Field | Description |
|---|---|
| Name | A short identifier for your project. Used in URLs and the sidebar. |
| Region | Available on paid plans. Pick the geographic region closest to where you want your data to live. |
| Product type | Defaults to Cloud Platform. Other types are for specialised use cases. |
Click Create. Provisioning takes a moment. When it completes you are dropped into the workspace.
Step 3. Open the workspace
The workspace is the main view for one sandbox. The sidebar shows tabs grouped by area:
- Workspace. Chat, Code, Preview.
- Database. Tables, SQL, Bin, Settings.
- Observability. Health, Gates, Development logs, ZeroClaw.
- Settings. Context, Secrets, Security.
You can keep tabs open across sessions. Each tab is independent.
Step 4. Authenticate an agent
The first time you use the chat tab, you choose which agent to run. The session picker offers:
| Session | What it runs |
|---|---|
| Shell | A shell with no AI. Useful for direct commands. |
| ZeroClaw | The autonomous agent runtime. |
| OpenCode | The OpenCode CLI. |
| Claude | Claude Code. |
| Codex | OpenAI's Codex CLI. |
| Cursor | The Cursor agent. |
If you pick Claude, Codex, or Cursor, the chat panel walks you through that agent's authentication flow. The agent prints a login URL or device code; you complete the flow in another tab and the credentials are stored on your encrypted volume.
You only authenticate each agent once per sandbox. Credentials persist.
Step 5. Send a message
Type a request and send it. For a first run try something simple:
Build me a Next.js app that displays the current time and updates every second.
The agent works inside the sandbox. It scaffolds the app, edits files, starts a preview server, and shows you the URL. Click it to view your app.
While the agent works:
- The Code tab shows files appearing in real time.
- The Preview tab shows the running app.
- The Observability > ZeroClaw tab shows the agent's process state.
Step 6. Approve a permission gate
If the agent asks to do something privileged (push to git, modify a secret, run a database write, deploy), a permission popup appears. Each gate has:
- A label and description of what is being requested.
- A time-to-live (how long the approval is valid).
- A risk indicator.
Approve to grant. Deny to refuse.
For routine requests you can configure standing rules under Settings > Security: ask (the default), allow always, or never.
Step 7. Review the audit log
Open Settings > Security. Every action is recorded: your login, the sandbox creation, the agent's gate requests, the deploy. Each entry is hash-chained so retroactive tampering is detectable.
Where to go next
Configure permissions
Set sensible defaults so the agent does not interrupt for routine tasks. Permission Gates.