🌐 Tavily Web Search

Information Retrieval · Data Aggregation · Search Engine

A commercial-grade search pipeline purpose-optimized for LLM agents. Through Tavily API's underlying calls, retrieve structured, noise-free quality data streams from the entire web — breaking through the LLM's training data cutoff limitations.

OpenClaw Team

🚀 Quick Install

Run the following command in your terminal to install:

npx clawhub install tavily-search

📊 Stats Overview

⭐ Stars	☁️ Total Downloads	👥 Active Users	🎯 Stable Version
821	94.5k	2,110	v1.2.0

🎛️ How It Works

Regular web requests or simple Google scraping contain massive ads and chaotically nested DOM elements, making it difficult for AI to extract information and causing extreme token consumption. The purpose-built Tavily integration solves this:

🔍 AI-driven Web Search Paradigm: Calls the industry-leading Tavily API. It's not just a query component — it adds cleaning nodes at the network layer, returning high-dimensional minimal JSON tree structures with context to local.
📝 Precise Full-text Extraction: Brute-force strips all anti-scraping components and sidebar ad streams from matched result pages, directly feeding articles, reports, and key content into the local LLM's context cache.
🛡️ Domain Precision Whitelist Filtering: Uses the included command pipeline to strictly limit result sources (include-domains / exclude-domains), preventing the model from ingesting non-compliant or heavily biased wiki sites.
🚀 Smart Search Mode Scaling: Built-in mode switching for different compute budgets. Use basic for high-speed bandwidth, or enable advanced to burn longer execution chains for deep-web resource mining.

🧭 Typical Use Cases

📑 Scenario 1: Deep Research Data Acquisition

When a user asks: "Please give me an in-depth analysis of yesterday's Apple Vision Pro market reception, excluding marketing accounts." The agent immediately engages this plugin's advanced deep mode with topic news (news-focused search) channel. The component automatically avoids all irrelevant entries, crosses multiple renowned tech news sources, and assembles a structured data matrix report with specific review paragraphs for the core analysis node.

🚨 Scenario 2: Development Dependency Live Debugging

If the model encounters a latest version library bug (like a sudden edge case in Next.js 14) that its internal knowledge lacks a solution for — leveraging the agent's dynamic reflection mechanism, it searches via Tavily's targeted channel with exclude-domains stackoverflow.com (excluding known answer overflow, seeking official voices), directly scanning official issue trackers and latest commit annotations to extract unpolluted first-hand fix methods.

💻 Command Reference

After installation, you can let AI call these autonomously via conversation, or manually trigger operations from the CLI:

Execute a news-oriented search query with deep mode and result limits:

clawhub execute tavily-search --topic news --max-results 10 "Latest AI policy regulations in EU"

Enable strict domain restriction scoping:

clawhub execute tavily-search --include-domains "github.com, arxiv.org" "Transformer mathematical explanation"

Switch to lightweight high-speed basic mode:

clawhub execute tavily-search advanced=false "Local weather mechanism in Tokyo"

🛡️ Requirements & Authentication

🔑 Tavily API Key Injection: Beyond npx package installation, you must register at tavily.com to obtain a free-tier or commercial license API key. Pre-load the environment variable before activating the Agent: export TAVILY_API_KEY="tvly-xxxxxxxxxxxx".
💰 Rate Limiting: Set hard limits on the cloud service side to prevent massive query bills caused by LLM infinite-loop scheduling.

🔗 View Source on GitHub