← all repositories
Johell1NS/browser-search

Your AI Agent Finally Got a Browser That Doesn't Get Banned

A self-hosted skill that lets AI agents search the web, browse results, and bypass anti-bot protections without hallucinating or paying for APIs.

518 stars JavaScript Coding AssistantsAgentsRAG · Search
browser-search
Collecting fresh signals — velocity needs a few days of history.
star history

What it does

browser-search is a drop-in skill for AI agents—OpenCode, Claude Code, Cursor, and others—that orchestrates three self-hosted tools to research the web end-to-end. It routes queries through SearXNG for metasearch, Camofox for standard browsing, and automatically escalates to CloakBrowser when sites throw up Cloudflare or Akamai challenges. The entire stack runs locally, costs nothing, and forces the agent to verify claims against live sources before answering.

The interesting bit

The anti-hallucination mechanism is architectural, not prompt-engineered. The SKILL.md enforces deterministic scripts and a rigid “search first, answer second” workflow, so even cheap models can’t freestyle commands or misinterpret outputs. It’s essentially a contract between the agent and the browser that the model isn’t allowed to break.

Key highlights

  • Three-tier escalation: SearXNG for fast metasearch, Camofox for everyday sites, and CloakBrowser for anti-bot fortresses—chosen automatically by the agent.
  • Fully self-hosted and free: no API keys, subscriptions, or rate limits; the README claims it runs comfortably on a Raspberry Pi.
  • Deterministic execution: natural language descriptions map to rigid scripts, preventing the model from getting commands wrong or hallucinating outputs.
  • Deep Research mode forces cross-referencing live web sources before answering, with automatic navigation and challenge detection.
  • Customizable SKILL.md: plain-text rules that can be edited, extended, or adapted to different agent environments.

Caveats

  • CloakBrowser has a known Windows bug where the browser may exit roughly 10 seconds after launch with a Pro license; the README suggests a workaround flag.
  • This is fundamentally integration glue around SearXNG, Camofox, and CloakBrowser—valuable orchestration, but not a from-scratch engine.

Verdict

Ideal for developers running local AI agents that need current, verified web data without feeding tokens to paid search APIs. Skip it if you already have a browsing stack you trust or balk at self-hosting three separate services.

Frequently asked

What is Johell1NS/browser-search?
A self-hosted skill that lets AI agents search the web, browse results, and bypass anti-bot protections without hallucinating or paying for APIs.
Is browser-search open source?
Yes — Johell1NS/browser-search is open source, released under the MIT license.
What language is browser-search written in?
Johell1NS/browser-search is primarily written in JavaScript.
How popular is browser-search?
Johell1NS/browser-search has 518 stars on GitHub.
Where can I find browser-search?
Johell1NS/browser-search is on GitHub at https://github.com/Johell1NS/browser-search.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.