Web Search Plus Plugin V2

Web Search Plus Plugin

Use this OpenClaw plugin for current web search, multi-provider research, URL extraction, routing configuration, provider health, and bounded extraction benchmarks. web_answer_plus is not part of the source-only surface; use Search plus Extract.

Enable the optional tools

The plugin registers all five tools with optional: true:

• web_search_plus
• web_extract_plus
• web_routing_config_plus
• web_search_health_plus
• web_extract_benchmark_plus

They must be permitted by OpenClaw's tool policy. Add the exact tools the agent needs without replacing its existing profile:

{
  "tools": {
    "alsoAllow": [
      "web_search_plus",
      "web_extract_plus",
      "web_routing_config_plus",
      "web_search_health_plus",
      "web_extract_benchmark_plus"
    ]
  }
}

For least privilege, remove any names the agent does not need. tools.alsoAllow is additive. tools.allow is the restrictive alternative and must also retain every non-plugin tool the agent needs. A configured plugins.allow is separate and must include web-search-plus-plugin-v2 so the plugin can load.

Good starter setup

Start with one or more of You.com, Serper, and Linkup. Tavily is first in the default extraction order:

Tavily → Exa → Linkup → Parallel → Firecrawl → You.com → Keenable → Serper → DonSeTch.

DonSeTch appears last in the stable extraction priority but is skipped by automatic routing until an operator explicitly enables its auto_allow gate. Extraction targets reject private/internal destinations by default. Inline content uses deterministic URL/context limits, and full-text continuation references live only in the process-local LRU until eviction or host restart.

Use spans=true for deterministic, query-conditioned passages. A query-matching ATX Markdown heading retains its body and deeper subheadings through the next same-or-shallower heading. At most two heading sections are added, each capped at 1,200 Unicode codepoints. Offsets are NFC Unicode-codepoint half-open ranges; within_preview reports whether a span is present in the inline preview.

Provider configuration

Hosted Search credentials/settings:

• serperApiKey, braveApiKey, tavilyApiKey, exaApiKey
• queritApiKey, linkupApiKey, firecrawlApiKey
• parallelApiKey, serpbaseApiKey, youApiKey
• searxngInstanceUrl, keenableApiKey
• monidApiKey for Octen via Monid
• tinyfishApiKey

Local provider:

• donsetchBin — absolute path to a separately installed DonSeTch executable

Important controls:

• parallelMode: turbo, fast (default), basic, or advanced; Parallel uses its stable v1 Search endpoint and is auto-allowed when configured
• octenTimeoutSeconds, tinyfishTimeoutSeconds
• donsetchTimeoutSeconds, donsetchMaxContentChars, donsetchTier
• keenableAllowPublic, searxngAllowPrivate, extractAllowPrivateUrls
• extractCharLimit, extractMaxUrls, extractMaxContextChars, extractDeadlineSeconds
• extractCacheMaxEntries, extractCacheMaxChars
• localeCountry, localeLanguage
• parallelMaxCharsPerResult, parallelMaxCharsTotal
• qualityBlockedDomains, qualityAllowedDomains, qualityDiversityRerank

OpenClaw plugin config is the only credential source; this package does not discover .env files.

Provider boundaries

• Exa exposes source-only neural /search. Unified day|week|month|year freshness becomes absolute UTC publication-date bounds. Deep, deep-reasoning, answer, and synthesis output are not exposed.
• Parallel uses stable v1 Search with fast as the default mode and participates in normal automatic routing when configured.
• Octen via Monid and TinyFish are Search-only and explicit-only by default. Both support freshness and domain filters; TinyFish also supports native news and locale.
• TinyFish's standard Terms permit Customer Data, including queries, to be used for model training and fine-tuning. Explicit-only is not a privacy guarantee. Review its Privacy Policy and the terms applicable to your account before sending sensitive data.