dsh-web-automation

Giving a local model real web access usually meant paying for a search API or handing queries to a third-party data broker, making it hard to balance privacy and cost. dsh-web-automation bundles a full local-first web stack into a single installable plugin: by default it talks only to DuckDuckGo and Bing public endpoints, requiring zero API keys and zero cloud dependencies, so inference and data stay on your machine.

Core capabilities span four areas: multi-engine search with parallel or fallback routing, BM25 enrichment, cooldowns, and graceful degradation; SQLite-backed cached fetching with ETag and Last-Modified revalidation so repeat requests within the TTL make no network call; platform-specific search across GitHub, Reddit, YouTube, Bilibili, RSS, and more, extending without code through config files or hot-reloaded versioned rule packs; and local-only history, stats, and cache-management tools. An optional Playwright-based browser automation sub-package adds open, navigate, snapshot, click, type, and screenshot operations. The security layer includes an SSRF guard (blocking loopback, private, and link-local targets, checking for DNS rebinding across every redirect hop) and a fail-closed browser approval flow, while LRU eviction keeps the local store bounded over time.

This plugin suits developers who deploy large language models locally and have hard requirements on data sovereignty and privacy. If you would rather not pay an API bill for every search, or want every outbound network call your model makes to be transparent, controllable, and auditable, this plugin delivers a ready-to-use local-first solution with no extra configuration required.