browser - Naïve

const client = naive.forUser(alice.id);

// Drive a session step by step
const s = await client.browser.createSession({ allowed_domains: ["example.com"], allow_writes: true });
const nav = await client.browser.navigate(s.session_id, "https://example.com"); // { url, title, ... }
const act = await client.browser.act(s.session_id, "click the first result");   // { success, action_taken, url }
const data = await client.browser.extract(s.session_id, "the product names and prices");

// Read real link targets + the current URL (what `extract` can't see — see note below)
const { url, links } = await client.browser.links(s.session_id, { contains: "linkedin.com" });
// → url: "https://example.com/…",  links: [{ text: "LinkedIn", href: "https://linkedin.com/in/…" }]

await client.browser.closeSession(s.session_id);

// Autonomous signup / login — the password is generated + vaulted server-side
const res = await client.browser.signup({ service: "figma.com", url: "https://www.figma.com/signup" });
if (isPendingApproval(res)) {
  // gated by the AccountKit — a human must approve before it runs
} 
await client.browser.login({ service: "figma.com", url: "https://www.figma.com/login" });

// List a user's sessions
const { sessions } = await client.browser.listSessions();

extract reads the page’s accessibility tree (visible text) — it does NOT return link href attributes or the page URL. Ask extract for a URL and you’ll get visible text or an element index, not the real link. To get link targets or the current URL:

links(session_id, { contains?, limit? }) — returns { url, links: [{ text, href }] } from a direct DOM read (the reliable way to capture LinkedIn/GitHub/profile URLs).
the url field returned by act and navigate — the page URL after the action, which is how you learn where a click navigated (including SPA route changes with opaque tokens).

act and navigate send a stable Idempotency-Key automatically. signup is sensitive and may resolve to a PendingApproval — check with isPendingApproval(res). Credentials never reach the model; the password is filled server-side via variable substitution. links is read-only and, on a logged-in (context-backed) session, is capability-gated + redacted exactly like extract. Human-only surfaces (live view, grants/revoke) live in the dashboard, not the SDK. See Browser.