dondai44423/donsetch

What it solves

DonSeTch provides AI agents with a high-performance, stealthy way to perform web research. It solves the problem of bot detection and "bot walls" (like CAPTCHAs) that typically block AI agents from accessing real-time web data, while also optimizing the token budget used by agents when processing web content.

How it works

DonSeTch is a single Rust binary that acts as a bridge between an AI agent and the web. It uses several advanced techniques to avoid detection:

• Native Chrome TLS: It drives Chrome's BoringSSL natively so its network fingerprints match a real browser.
• Solve-and-Bounce: It uses a browser to solve challenges and then hands the cookies back to a faster, lightweight fetcher.
• Keyless Search: It aggregates results from over 10 search backends in parallel without requiring API keys.
• Self-Improving Fetch: It learns from every request to optimize routing and cookie lifetimes per domain.
• Agent-First Optimizations: It uses "reference handles" (short IDs instead of long URLs) and "probe mode" (verifying claims without fetching the whole page) to save tokens.

Who it’s for

It is designed for AI agent developers and users who need "one-shot" rapid-fire research (searching, reading a few pages, or verifying a PDF) rather than bulk data harvesting or mass scraping.

Highlights

• MCP Compatible: Works with Model Context Protocol clients like Claude Code, Cursor, and OpenCode.
• Pixel-Fusion PDF: Fuses glyphs and rendered pixels for high-fidelity PDF reading, including auto-OCR for scanned documents.
• Temporal Stealth: Implements TLS session resumption and connection pooling to mimic human behavior.
• Keyless Search: Provides free search capabilities via cross-engine consensus and semantic reranking.
• Resurrection Fetch: Automatically serves Wayback Machine snapshots if a link is dead.