Browser Use
This example wires together Browser Use and Inkbox so your agent has a real email address and a live browser in the same loop. The agent can navigate the web, sign up accounts, fill forms, extract content, and send or receive email, all from a single natural-language prompt.
Setup
Requires uv.
| Variable | Required | Description |
|---|---|---|
INKBOX_API_KEY | yes | Your Inkbox API key from console.inkbox.ai |
BROWSER_USE_API_KEY | yes (cloud mode) | Your Browser Use API key from browser-use.com |
OPENAI_API_KEY | yes | OpenAI API key for the Browser Use LLM |
Quickstart
On first run you'll be prompted to pick or create an Inkbox identity. Chrome will be auto-launched in debug mode if it isn't already running.
CLI flags
| Flag | Default | Description |
|---|---|---|
--env | local | local uses your local Chrome; cloud uses Browser Use Cloud (no local Chrome needed) |
--chrome-debug-url | http://127.0.0.1:9222 | Chrome remote debugging URL (local mode only) |
--keep-browser | off | Keep the browser tab open after the task completes |
How it works
- You pick (or create) an Inkbox identity — the agent's handle and email address
- Chrome launches automatically in debug mode, or a Browser Use Cloud session is created
- The agent loads a system prompt from
SKILL.mdwith the identity's handle and email interpolated - Browser Use runs the agent loop: the LLM uses browser actions (navigate, click, fill, screenshot) and Inkbox email tools to complete the task
- Browser and Chrome are cleaned up when done (unless
--keep-browser)
Email tools
The following Inkbox email tools are registered on the Browser Use controller and available to the agent during the run.
| Tool | Description |
|---|---|
send_email | Send an email with optional reply threading, CC, BCC, and HTML body |
list_emails | List recent emails, optionally filtered by inbound or outbound |
check_unread_emails | List unread emails only |
mark_emails_read | Mark specific emails as read by message ID |
read_email | Read a specific email in full, including body text |
get_thread | Retrieve all messages in an email thread by thread ID |
Browser actions (navigate, click, fill, extract, screenshot) are handled automatically by Browser Use.
Local Chrome setup
When using --env local (the default), Chrome must be running with remote debugging enabled. The CLI will auto-launch Chrome for you. To launch it manually: