web-search-plus-plugin-v2

name	web-search-plus-plugin-v2
version	3.2.0
description	OpenClaw plugin for Routing v2 multi-provider web search, research mode, canonical-source reranking, spam/mirror filtering, unified freshness and news vertical, locale defaults, Tavily-first extraction, quality reports, onboarding CLI, and runtime routing preferences. Registers `web_search_plus`, `web_extract_plus`, and `web_routing_config_plus`.

Web Search Plus Plugin

Native OpenClaw plugin that gives agents one clean web surface: search, extract, and routing config. web_answer_plus is intentionally removed in v3.0.0; use search plus extraction instead.

Tools

• web_search_plus
• web_extract_plus
• web_routing_config_plus

Good starter setup

Recommended starter preset:

• You.com
• Serper
• Linkup

Run:

web-search-plus-setup setup --preset starter --config ./web-search-plus-plugin.config.json

Tavily is the default first extraction provider in auto mode. The fallback chain is Tavily → Exa → Linkup → Parallel → Firecrawl → You.com → Keenable → Serper (webpage scraper). Extraction targets are validated against private/internal destinations by default; oversized pages return a head/tail window governed by extractCharLimit.

Config fields

Search providers:

• serperApiKey
• braveApiKey
• tavilyApiKey
• exaApiKey
• queritApiKey
• linkupApiKey
• firecrawlApiKey
• parallelApiKey
• serpbaseApiKey
• perplexityApiKey
• kilocodeApiKey
• youApiKey
• searxngInstanceUrl
• keenableApiKey

Extra settings:

• braveSafesearch
• searxngAllowPrivate
• routingConfigPath (namespace only; runtime prefs are in-memory)
• keenableAllowPublic (opt-in keyless Keenable public tier)
• extractAllowPrivateUrls (opt-in private/internal extraction targets)
• extractCharLimit (inline extract budget, default 15000)
• localeCountry / localeLanguage (default search locale; "auto" language enables query inference)
• parallelMaxCharsPerResult / parallelMaxCharsTotal (Parallel extract budgets)
• qualityBlockedDomains / qualityAllowedDomains (spam/mirror blocklist overrides)

Routing v2

Auto routing is class-aware and benchmark-backed. Key classes: multilingual/current, local/shopping, docs/api, academic/arxiv, community/reddit, security/cve, official/vendor-release, official/regulatory, finance/IR, weather/factual, oss-discovery, and answer/synthesis.

Default auto pool: You.com, Serper, Exa, Firecrawl, Tavily, Linkup. Guarded providers require auto_allow[provider]=true for auto routing: Brave, SerpBase, Querit, Parallel, Perplexity, Kilo Perplexity.

Every search routing object exposes language_hint, routing_class, and routing_policy. Pass quality_report: true for provider scores, result quality hints, fallback diagnostics, and authority_signals (canonical/demoted domain hits and primary-source top-result flag). Recent provider latency/success behavior feeds bounded adaptive score adjustments (routing.adaptive_adjustments).

Known SEO mirror/scraper domains are filtered from results and one domain is capped at two head slots (metadata.result_filter); site:/include_domains constraints bypass both. freshness: day|week|month|year applies native recency filters where supported (metadata.freshness), search_type: news uses Serper's native news vertical (metadata.search_type), and localeCountry/localeLanguage steer provider locale (metadata.locale).

For canonical-source classes (official/vendor-release, docs/api, official/regulatory, finance/IR, security/cve), auto-routed results are reranked so primary sources outrank mirrors; reorderings are reported in metadata.intent_rerank.

Research mode

web_search_plus(mode="research") queries up to 3 providers concurrently, deduplicates across them, and extracts the top research_extract_count URLs (default 3) into source_summaries for grounding. Use research_providers to pick providers explicitly and research_time_budget (seconds, default 55) to cap wall-clock cost. Failures are best-effort diagnostics, not hard errors.

Usage guidance

Prefer web_search_plus for live/current info, prices, weather, sports, schedules, and finding raw sources. Use mode="research" for grounding-heavy questions that benefit from multiple providers plus extracted full text. Use web_extract_plus once URLs are known.

OpenClaw plugin config remains the source of truth for credentials; runtime code does not rely on direct .env reads.

Perplexity provider split:

• perplexity → direct Perplexity API (perplexityApiKey)
• kilo-perplexity / kilo_perplexity → Kilo gateway (kilocodeApiKey)