Modern AI agents often need access to a user's real web context — logged-in pages, internal sites, and browser state — to complete multi-step workflows or gather evidence. This project provides a local bridge that safely exposes those capabilities to any shell-capable agent while keeping automation visible and interruptible by the user.
What Sets It Apart
- Reuse of real login state: agents operate using the browser profile's existing cookies and sessions, enabling access to pages the user already has permission to view (so agents can interact with internal dashboards, SaaS UIs, or single-sign-on flows without test accounts).
- Visible, isolated Agent Window: automation runs in a separate, user-visible window and can explicitly borrow and return existing tabs, preventing silent changes to the user's working windows.
- CLI + extension architecture: a local
bskCLI/daemon plus a Chromium extension mediates control; this design supports any agent that can call the CLI and integrates with DeepSeek Harness via an official plugin for richer in-UI tools. - Action primitives and evidence: snapshot/observe/read primitives, click/fill/press, navigation, screenshots, and request/console debugging link actions to network/console evidence for reproducible debugging or audits.
Who It's For and Trade-offs
Great fit if you need agents to perform real-world web workflows that require the user's authenticated browser context (e.g., form submissions, internal site checks, UI smoke tests, evidence collection). It is particularly useful when you want automation to remain visible and optionally human-interruptible during captcha, login, or confirmation steps. Look elsewhere if you require headless/cloud-only browser automation without local user interaction, or if you must avoid any local process that can access browser profiles; this approach deliberately uses local browser state and a local daemon. It also requires installing a browser extension and a local CLI, so fully managed/cloud-hosted integrations are not the target.
Where It Fits
Acts as a local integration layer between general-purpose AI agents (or harnesses like DeepSeek Harness) and a user's Chromium-based browser, filling a niche between pure headless automation frameworks and remote/cloud agent services by prioritizing authenticated, user-visible automation with auditability.
