Skip to content
Dashboard
Getting Started

How Lightcone Works

Understand Lightcone's architecture — cloud browser sessions, agent layers, and how they fit together.

Lightcone is a cloud platform that gives your code the ability to control browsers and desktops. Every browser runs in an isolated container in the cloud — you never install or manage browsers locally.

┌──────────────────────────────────────────────────────────┐
│ Your Code │
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌───────────────┐ │
│ │ Computers │ │ Agent Tasks │ │ Responses API │ │
│ │ (low-level) │ │ (managed) │ │ (CUA) │ │
│ └──────┬───────┘ └──────┬───────┘ └───────┬───────┘ │
└─────────┼─────────────────┼──────────────────┼───────────┘
│ │ │
▼ ▼ ▼
┌──────────────────────────────────────────────────────────┐
│ Lightcone API │
│ api.tzafon.ai │
├──────────────────────────────────────────────────────────┤
│ │
│ ┌────────────┐ ┌────────────┐ ┌────────────┐ │
│ │ Browser │ │ Browser │ │ Desktop │ ... │
│ │ Session │ │ Session │ │ Session │ │
│ │ (Chromium) │ │ (Chromium) │ │ (Linux) │ │
│ └────────────┘ └────────────┘ └────────────┘ │
│ │
│ Sandboxed containers · Stealth mode · Proxy rotation │
└──────────────────────────────────────────────────────────┘

Lightcone offers three interfaces at different levels of abstraction. All three create the same underlying browser sessions — they differ in how much control you have.

You create a browser session and send individual actions: click, type, navigate, screenshot. You decide what to do at every step.

with client.computer.create(kind="browser") as computer:
computer.navigate("https://example.com")
computer.click(400, 300)
computer.type("hello")
result = computer.screenshot()

Best for: browser automation, web scraping, testing, and any workflow where you know exactly what steps to take.

Learn more →

You give a natural language instruction, and an autonomous AI agent drives the browser to complete it. The agent sees the screen, decides what to click, and acts on its own.

for event in client.agent.tasks.start_stream(
instruction="Go to amazon.com and find the price of the MacBook Air M4",
kind="browser",
):
print(event)

Best for: tasks that are easy to describe but tedious to script — research, data collection, form filling across unfamiliar UIs.

Learn more →

You build a computer-use agent (CUA) loop: the model looks at a screenshot, tells you what action to perform, you execute it, and feed back the new screenshot. You control the loop and can inject custom logic between steps.

response = client.responses.create(
model="tzafon.northstar-cua-fast",
input="Search Wikipedia for 'Alan Turing'", # or an array with screenshots
tools=[{"type": "computer_use", "environment": "browser"}],
)
# Execute the action, screenshot, send back, repeat

Best for: custom agent workflows where you need to validate results, branch on conditions, or integrate with other systems between steps.

Learn more →

When you call client.computers.create(kind="browser"):

  1. Provision — Lightcone spins up an isolated container with a Chromium browser (or full Linux desktop for kind="desktop")
  2. Connect — You get back a session ID and endpoints (API, CDP, WebSocket)
  3. Interact — You send actions via the API. Each action executes in the browser and returns a result
  4. Terminate — The session is destroyed, freeing resources. If persistent=True, cookies and local storage are saved for reuse

Sessions are fully isolated from each other. Each session runs in its own container with its own filesystem, network stack, and browser profile.

Lightcone’s browsers are configured to avoid detection by anti-bot systems:

  • Browser fingerprinting — realistic browser fingerprints that match consumer browsers
  • WebDriver flags — automation indicators are suppressed
  • Residential proxies — optional routing through residential IP addresses (use_advanced_proxy=True)
  • TLS fingerprinting — TLS handshake matches real Chrome

This means you can automate websites that block tools like Selenium, Puppeteer, or plain HTTP requests.

Lightcone provides AI models for computer-use and text generation:

ModelPurpose
tzafon.northstar-cua-fastComputer-use tasks — optimized for seeing screenshots and deciding actions
tzafon.sm-1General text generation, chat, and tool calling

Both models are available through the Responses API and Chat Completions API. The Agent Tasks API uses northstar-cua-fast by default.

  • Quickstart — create your first browser session in 5 minutes
  • Computers — full reference for browser and desktop sessions
  • Integrations — connect to LangChain, Playwright, OpenAI CUA, and more