AI agent
Browser context for the Agent
Open pages in Kition’s embedded desktop browser, sign in manually, and let the Agent read supported page context without promising full browser automation.
Updated
By Kition Docs
Verified against Kition 0.1.41 (d3e1b931) on 2026-08-29
What ships in 0.1.41
The desktop app includes an embedded browser surface with navigation, back, forward, reload, stop, persistent browser partitions, page-context extraction, and a Browser Sites list.
The Agent can work with active browser context and can request that Kition open a page. The extracted context can include visible text, HTML snapshots, links, content blocks, page metadata, and structured entities.
Open a page and continue a task
For an explicit web target, the client can preflight the browser, extract usable context, and continue the original request. Browser auto-continuation is bounded to three attempts before the task is reported as blocked.
- Open a browser tab from the workspace and navigate to the target site
- Sign in manually when the site requires authentication
- Ask the Agent to inspect, summarize, compare, or extract information from the current page
- Include an explicit URL and a browser/research action when you want Kition to auto-open a target
- Review the extracted facts against the visible page before writing them into documents or tables
Automation boundary
The 0.1.41 public client does not expose a general Agent API for arbitrary DOM clicks, typing, form submission, JavaScript execution, screenshots, PDF export, iframe control, or selector-based action scripts.
User navigation and login remain separate from Agent context reading. Do not document the embedded browser as a full Playwright-style automation tool in this release.
Cookies and Browser Sites
Browser sessions use persistent Electron partitions per profile, so a manually established login can remain available for later visits. The Browser Sites panel records host, profile, last visit, and a probed login status; it does not copy cookie values into the registry.
Forget site clears matching cookies and origin storage for the selected site/profile. Use it after testing a sensitive service or when a login should no longer persist.
Safety
- Use a separate low-privilege account for sites the Agent will inspect
- Do not open banking, production administration, or highly sensitive sessions unless the task requires it
- Treat page text as untrusted input that can contain prompt injection
- Remember that extracted page context may be sent to the selected model provider
- Use Forget site and remote sign-out when a persistent login is no longer needed
When context is empty
- Wait for the page to finish loading, then retry
- Confirm the correct tab is active
- Sign in manually and refresh the login status
- Reload the page if the extracted HTML or visible text is stale
- Treat an unavailable page as a blocker; do not let the Agent fabricate results
Implementation sources
- v0.1.41: src/lib/productFeatures.ts
- v0.1.41: src/features/agent/lib/agentBrowserIntent.ts
- v0.1.41: src/features/agent/lib/agentBrowserContinuation.ts
- v0.1.41: src/features/agent/lib/agentBrowserPreflight.ts
- v0.1.41: electron/generic-browser-session.mjs
- v0.1.41: src/features/workspace/components/WorkspaceBrowserSitesPanel.tsx
Related pages
Agent chat basics
Open a workspace-scoped chat, attach only the context you need, choose a model, and review the streamed work.
Agent tool calls
Read the actual runtime tool names, availability signals, inputs, outputs, errors, and changed-file results shown in the Agent timeline.
Agent approvals and safety boundaries
Use the shell-command approval card and feature-specific previews without assuming Kition 0.1.41 has a universal per-tool permission editor.
