tavily

What This Skill Does

The Tavily AI Search skill provides OpenClaw agents with advanced, AI-optimized web search capabilities. Unlike standard search engine APIs that return raw HTML or fragmented snippets, Tavily is engineered specifically for Large Language Model consumption. It retrieves clean, structured, and context-rich data, allowing your agents to perform deep research, verify facts, and synthesize information from across the web. Whether you need a lightning-fast basic query for quick facts or a comprehensive, multi-step search for complex research, this skill handles the heavy lifting of content extraction and source filtering, ensuring that your agent receives the most relevant, noise-free information available.

Installation

To integrate this skill into your local OpenClaw environment, run the following command in your terminal:

clawhub install openclaw/skills/skills/bobdevibecoder/bobagent-tavily

Once the installation completes, the skill will be available for all active agents to leverage via their standard tool invocation interface.

Use Cases

• Comprehensive Research: Deep dive into niche topics where accuracy and authoritative sources are paramount.
• Fact-Checking: Verify current events or technical documentation against verified, live online sources.
• Trend Analysis: Utilize the news-focused topic filter to keep your agents informed of developments within the last seven days.
• Content Curation: Aggregate high-quality links and metadata for summary reports or AI-generated briefings.
• Filtered Searches: Restrict agent knowledge retrieval to specific, trusted documentation domains (e.g., official docs or academic repositories) to prevent hallucinations or low-quality data ingestion.

Example Prompts

1. "Tavily, please research the current state of decentralized finance (DeFi) regulations in the European Union and summarize the key challenges for developers."
2. "Perform an advanced search for the latest breakthroughs in cold fusion technology from the last week and provide a bulleted list of credible news sources."
3. "Search for Python best practices regarding asynchronous programming, ensuring you only retrieve results from python.org and realpython.com."

Tips & Limitations

• Mode Selection: Use 'basic' mode for latency-sensitive tasks where speed is critical, and reserve 'advanced' mode for research tasks that require exhaustive data gathering.
• Domain Filtering: Leverage include/exclude domains proactively. This significantly improves the signal-to-noise ratio and helps the agent focus on verifiable, high-authority information.
• Images: You can request image URLs alongside text results if your agent is processing multimodal data or visual explanations.
• Latency: Keep in mind that 'advanced' searches take 5-10 seconds. Avoid putting these inside tight loops in your agent's decision-making flow to prevent timeouts.