Tilion creates an isolated browser for you over a REST call. You drive it with the tools you already use and pay only while it runs.
The response contains a connect_url you can hand straight to Playwright. Start with the Quickstart.

​
What you get

Every session is a full Chromium with site isolation intact, driven over the standard Chrome DevTools Protocol. Each session gets its own browser, destroyed on release, and no cookies, profile, or disk state is ever carried between sessions or between customers. Your existing tooling works unchanged. Playwright, Puppeteer, chromedp, and raw WebSockets all attach the same way, so there is no Tilion SDK to learn and no rewrite: point connectOverCDP at the URL and your existing script runs. Paid plans can route traffic through residential IPs, configured per session.

​
Two ways to use it

You can drive the browser yourself. Create a session, attach over CDP, and do anything Chromium can do. This suits multi-step flows, logins, and anything stateful.
Or you can ask for a page. You still create a session, but a single fetch call returns the content and there is no CDP client to run.

​
Documentation

MCP
  • The MCP overview connects any MCP client to a real cloud browser it can drive.
  • The Tools reference documents every tool an agent can call, including the input actions.
Sessions
  • The Quickstart gets a script working in about two minutes.
  • Sessions covers the lifecycle: create, read, drive, and release.
  • Connecting a browser covers CDP, connect tokens, reconnecting, and troubleshooting.
  • Session configuration lists every create option, including regions and personas.
  • Proxies covers residential egress and IP rotation.
Fetch and content
  • Fetch returns page content without driving a browser, and Async jobs covers polling the long-running form.
  • Extract returns the same content as clean markdown for a model context.
  • Search returns search-engine results as structured JSON.
  • Inspect reports the network and behaviour a page exhibits.
Agent & tasks
  • Agent tasks run named, pre-built browser flows.
  • Templates run stored flows by slug.
  • Skills turn a natural-language goal into a cached, replayable flow.
Live view & recordings
  • Live view covers watching a session in real time.
  • Replays covers recording a fetch and playing it back.
Account and reference
  • Usage and credits covers month-to-date usage and the credit balance.
  • Limits explains plans, concurrency, and quotas.
  • Errors lists every status code, body shape, and retry rule.
  • For AI agents collects llms.txt, the OpenAPI schema, and the rules a coding agent should encode.

​
Before you build

One design detail surprises people, so it is stated up front:
connect_url is minted on demand. It is returned by POST /v1/session and GET /v1/session/{id}/connect only. Reading a session back returns connect_url: null by design; the session is fine. Tokens expire after 300 seconds and cannot be cached.
The details are in Connecting a browser.