DonSeTch

The web, for AI agents.

Why · Demo · Sponsors · Install · Quickstart · The 4 tools · Fetch · Search · PDF · Stealth · Compare · Limits · CLI

---

DonSeTch gives any AI agent full web research from a single local process: fetch, search, crawl, screenshot. Four tools, zero API keys, zero accounts, Rust, one binary. The transport is built from scratch (no hyper, no Playwright, no Selenium), so the fetch tier is fast and stealthy, and the tool surface stays small enough to fit an agent's context.

Works with every MCP client (Claude Code, Cursor, OpenCode, Pi, Hermes) and as a standalone CLI.

🤝 Use Bright Data? Support DonSeTch. Their proxy, SERP and Web Unlocker products come through this partner link and part of it comes back to keep DonSeTch free. You pay nothing extra: https://get.brightdata.com/ivqwoicrrlbr Bright Data plugs into the tool itself: the bd SERP provider, the Web Unlocker tier-3 bypass and the unlocker key type. Bright Data setup guide: zones, API token, HTML/JSON and troubleshooting.

✨ Why it's different

	What it does
🛡️ Real Chrome TLS	Drives Chrome's own BoringSSL natively. Your ClientHello IS Chrome's ClientHello, ML-DSA signature algorithms included. Emergent from the real engine, not a faked table that rots.
⏱️ Temporal stealth	TLS session resumption, conditional revalidation (304), persistent cookies, connection pooling, TCP Fast Open. The loudest remaining bot tell, and nobody else fakes it.
👻 Solve-and-bounce	The browser solves the challenge and hands cookies back to tier 1, which then fetches at full speed. The browser almost never fetches content.
🧠 Self-improving fetch	Learns from every fetch: cookie lifetimes adapt, walls that beat a real browser twice go into cooldown, searches pre-solve known walls. Receipts in status and doctor --improve.
🔑 Keyless search	10+ backends in parallel, fused by cross-engine consensus plus local semantic reranking. No keys, $0 forever. BYOK optional.
📄 Pixel-fusion PDF	Glyphs and rendered pixels come from the same content stream and are fused deterministically, with a per-region trust audit. Scanned PDFs auto-OCR.
🧬 Built from scratch	Own HTTP/2 (HPACK, flow control, priority), own extraction engine, own PDF parser, own search aggregator, own crawl engine.
🔗 Token control	Links render as [text](L12), results as S1…Sn, and fetch S3 just works: 3 tokens instead of 80. focus, toc, section, must_contain and since_last each cut a page down to what the agent actually needs.
🪶 ~2.4k tool schema	All four tools: 2,369 tokens (o200k_base, compact tools/list JSON).
🩺 doctor	One command sweeps config, search health, egress, TLS, browser, DNS, captive portal, secret-store permissions, and fixes what is mechanically fixable.

🎬 Demo

(30-second walkthrough: search, bot-wall bypass, crawl)

(Pi agent session: live research with DonSeTch as a native extension)

💛 Our Sponsors