Show HN: Nxtscape – an open-source agentic browser

Concept and Scope of an “Agentic Browser”

• “Agentic browser” is interpreted as a browser where AI agents can navigate, click, fill forms, group tabs, and perform multi-step tasks on the user’s behalf.
• Some see this as the “missing piece” after tools like Dia and Claude Desktop: an assistant that acts directly in the pages you’re already using.
• Others argue the right abstraction is an OS-level agent plus webviews, not a standalone browser.

Browser Fork vs Extension / CDP

• Many question why this is a Chromium fork instead of a Chrome extension or an MCP tool using Chrome DevTools Protocol.
• Supporters of the fork point to: using the accessibility tree, integrating a bundled LLM, building an “AI-friendly DOM”, and tighter control of UX.
• Critics say almost all of this is achievable via extensions/CDP, and that forking Chromium brings a heavy ongoing security/maintenance burden.

Security, Privacy, and Local vs Cloud

• Strong concerns about:
  • Agents acting on highly sensitive sites (banks, health portals).
  • Prompt injection leading to destructive actions.
  • Bypassing browser security (e.g., trusted events, full‑screen, form writes).
• Suggestions include: per-site opt-in for read/write, human confirmation before irreversible actions, shadow profiles with limited sessions, rate limiting, and robust “undo/restore checkpoint”.
• Some users will only trust an agent that runs entirely locally; Ollama integration is welcomed but there’s suspicion that cloud APIs are the primary intended path.

robots.txt, Scraping, and Website Interests

• The project currently ignores robots.txt.
• One camp: AI that follows multiple links and bulk-summarizes data is effectively a scraper and should respect robots.txt, especially to avoid overloading or bypassing monetization.
• Opposing camp: this is a user agent, not a crawler; robots.txt is defined for recursive crawlers, not interactive tools acting on explicit user requests. Applying robots.txt to user agents risks allowlists and loss of browser freedom.
• Some propose a new standard (e.g., “ai-browsers.txt”) for AI-assisted browsing.

UX, Chat Interface, and Reliability

• Split views on chat as the primary interface:
  • Critics say no one wants to “chat” for productivity; agents should mostly act via learned “recipes” and structured actions, with chat as fallback.
  • Others report real productivity gains with chat-based tools (e.g., Dia, Cursor) and are happy to instruct an agent conversationally.
• Early UX issues noted: mode confusion (bouncing between “agent” and “productivity” modes), lack of an ungroup-tabs API, need for a good undo/checkpoint model, and better handling of ambiguous tasks.

Productivity, Tab Management, and Use Cases

• Some enthusiasm for:
  • Intelligent tab grouping and session management.
  • Automating repetitive web tasks (multi-site price comparison, CSV downloads, form-filling).
  • Acting as a personal “bodyguard” against user-hostile sites, clutter, and distraction.
• Others dismiss the “10x productivity” framing and view this as “AI slop” or a solution in search of a problem, especially when supervision overhead cancels any gains.

Licensing, Business Model, and Competition

• AGPL is polarizing:
  • Supporters like that it allows commercial use but prevents proprietary forks and “exploitative” models.
  • Detractors say it deters businesses from adopting or extending the code.
• Proposed monetization: open-source core with an enterprise edition (parallels drawn to Island, Chrome Enterprise, Brave).
• Multiple commenters doubt a small team can win against incumbents (Chrome with built-in LLM, other AI browsers, extension-based agents).

Branding, Naming, and Platform Support

• Heavy criticism of the fox logo and Netscape-like name; some expect trademark trouble and find the branding misleading for a Chromium fork.
• Many requests for Linux and Windows builds; some defend focusing on macOS first, others note an extension would reach far more users immediately.