When to use local vs cloud
Lightpanda runs the same engine two ways. You can run the binary yourself, on your own machine or infrastructure, or connect to Lightpanda Cloud and let a remote browser do the work. The engine is identical. What differs is where the browser runs, which interfaces you can reach, and who sees the traffic.
Choose local or cloud
Run local when you want full control and want nothing to leave your machine. The binary runs on your hardware, so pages you visit and data you extract stay in your infrastructure. Local gives you every interface:
- The CDP or the WebDriver BiDi server
- The MCP server, with all tools
- The built-in AI agent
You manage the process, the scaling, and the network egress yourself.
Use the cloud when you do not want to run or scale browsers. You sign up, get a token, and connect a remote browser with no process to manage and no host to keep alive. Lightpanda Cloud runs the browser on its servers and adds the parts a hosted service needs: a console to manage your account and tokens and review recent sessions, usage-based plans with concurrency and monthly limits, a managed proxy pool that helps when a target blocks datacenter IPs, and an HTTP API that returns a page’s HTML or Markdown from a single request with no CDP script. Because requests run on Lightpanda’s servers, your target URLs and the fetched content pass through Lightpanda’s infrastructure.
Confidentiality and telemetry
The main difference is where your traffic goes.
Running local, the browser runs on your machine. The pages you load, their content, and your cookies never reach Lightpanda. They only reach the sites you visit, and the proxy if you configure one. The binary does send anonymous usage telemetry to telemetry.lightpanda.io: whether a navigation used TLS, the kind of context (page, iframe, popup), and, for the agent, the LLM provider and model name. For tool calls it sends counts per tool, caller and outcome, with their total duration. The outcome is whether the call succeeded or which kind of error it hit, not the tool’s result. MCP sessions also report which known MCP client is connected. The outcomes and known clients are listed in telemetry.zig. It never includes tool arguments or results, the URLs you visit, or the page content. Telemetry is off in debug builds, and you can disable it in any build by setting the LIGHTPANDA_DISABLE_TELEMETRY environment variable. You can read Lightpanda’s privacy policy at: https://lightpanda.io/privacy-policy .
Running on the cloud, the browser runs on Lightpanda’s servers, so your target URLs and the pages they return pass through Lightpanda’s infrastructure. You can review your recent sessions in the dashboard . If your workload cannot leave your own environment, run local.
Feature coverage
This table shows which interfaces and features you can reach in each mode. Coverage is still expanding, so the cloud does not yet match local on every interface.
| Interface / feature | Local | Cloud |
|---|---|---|
CDP server (serve) | Yes | Yes |
| MCP server | Yes | Yes |
| AI agent | Yes | No |
fetch command | Yes | No |
| HTTP API | No | Yes |
| Proxies | Yes, your own proxy | Yes, shared pool with country selection, or your own |
For CDP, both modes speak the same protocol: the same Puppeteer, Playwright, and chromedp scripts run against a local endpoint or a cloud endpoint with your token. The HTTP API is the cloud counterpart to the local fetch command: both return one-shot HTML or Markdown with no CDP script. See Getting started to connect to the cloud.