dsh-web-tools

A unified multi-provider Web Runtime for DeepSeek Harness.

Preserves the native DSH web_search / web_fetch tool contracts while providing multi-provider aggregation, automatic fallback, multi-key pooling, quota and health monitoring, provider-native preference tuning, page content extraction, and per-session Search Mode at the host runtime layer.

English | 简体中文

Features

• Zero-Overhead SearchHints Semantic Layer: Built-in deterministic intent extraction directly from queries (coding, academic research, news freshness, and domain constraints like site:github.com or after:YYYY-MM-DD), intelligently mapping to native parameters and specialized indices without LLM planning overhead or latency.
• 8 Search Providers Deeply Adapted:
  • Parallel: Dual-layer semantics (objective soft-steering + clean search_queries) and source_policy domain/freshness filters.
  • Firecrawl: Coding/technical queries query the Developer Index (categories: ["developer"]), supporting research and tbs time filters.
  • Exa: Exact category mapping (publication / news / financial report), ISO-8601 date ranges, and domain constraints.
  • You.com: Native boost_domains soft-weighting, freshness presets, and geo/language targeting.
  • Brave Search: LLM Context endpoint with pd/pw/pm/py freshness filters, country, and search language.
  • Tavily: Full-tier chunks_per_source chunking control, news topic and time range filtering (gracefully falls back to general search for finance/code).
  • SearXNG: Self-hosted metasearch with categories (it/science/news) and time_range.
  • Jina: Query noise reduction and ReaderLM-v2 high-precision markdown extraction.
• Platform Search Sources (Xiaohongshu & Twitter / X):
  • Browser Bridge Architecture: Leverages user authenticated browser sessions via a lightweight MV3 extension, with 0 raw cookie storage on the Host process, bypassing fragile reverse-engineered APIs.
  • Xiaohongshu: Waterfall DOM incremental extraction preserving complete signed xsec_token URLs, note text, and engagement metrics.
  • Twitter / X: Native query operator mapping (from:, since:, lang:) and semantic DOM tweet extraction.
  • Graceful Web Fallback: Seamlessly falls back to site:xiaohongshu.com / (site:x.com OR site:twitter.com) when disconnected.
• Native DSH Tool Compatibility: No bespoke tools like web_search_exa; agents invoke standard web_search and web_fetch contracts seamlessly.
• Multi-Query Support: Handles DSH queries[] payloads concurrently across independent search dimensions.
• Multi-API-Key Pooling: Assigns keys per provider, balances concurrent requests by lowest in-flight count, and fails over across keys on authentication errors.
• Deterministic Provider Fallback: Automatically cascades through the fallback chain on network failures, timeouts, 5xx server errors, 429 rate limits, or exhausted quotas.
• 429 Retry-After Cooldown: Enforces zero-request cooldown windows when servers return Retry-After headers, skipping rate-limited providers immediately without redundant network overhead.
• Configurable Routing Policies: Supports Ordered, Round-Robin, and Random initial provider routing.
• One-Click Preference Presets: Offers Fast, Deep, Economy, and Recommended presets to quickly adjust execution parameters across providers.
• **Native Page

…