Skip to main content

Tavily

Empower your agents with real-time web knowledge using Tavily, the search engine built specifically for AI.

💡 Core Concepts​

To effectively use Tavily, you should understand how it interacts with your agent.

1. What is Tavily?​

Tavily is a search API optimized for LLMs. Unlike traditional search engines, it returns concise answers.

2. Configuration Parameters​

You can customize the search behavior in the config block:

  • max_results: Limit the number of search results returned (e.g., 5).
  • include_answer: Set to True to get a short, synthesized answer to the query.
  • include_raw_content: Set to True if you need the full text of the scraped pages.
  • search_depth: Choose between basic (faster) or advanced (deeper search).

⚙️ Configuration Steps​

Follow these steps to enable web search capabilities for your agents.

Get your API Key​

  1. Sign up at tavily.com.
  2. Create a new API Key in your dashboard.
  3. Copy the key (it starts with tvly-).

Hardcoding keys is risky. We recommend using the Key Vault.

  1. Navigate to the Key Vault section in SVAHNAR Platform.
  2. Add a new secret named TAVILY_API_KEY.
  3. Paste your Tavily API Key as the value.
tip

This allows you to reference the key in your YAML as ${TAVILY_API_KEY}.

Configure the Tool​

Add the tool to your agent's configuration. You must ensure the agent has a clear instruction to use it.

Minimal Configuration:

tools:
  tool_assigned:
    - name: Tavily
      config:
        API_key: ${TAVILY_API_KEY}

📚 Practical Recipes​

Recipe 1: Standard Web Search Agent​

Use Case: An agent that performs broad searches to answer general questions.

create_agent_network:
  agent-1:
    agent_name: "web_search_agent"
    LLM_config:
      params:
        model: "gpt-5-mini"
    tools:
      tool_assigned:
        - name: Tavily
          config:
            API_key: ${TAVILY_API_KEY}
            max_results: 5
    agent_function:
      - "Search the web for authoritative resources about the user's query."

Recipe 2: Advanced RAG Fetcher​

Use Case: An agent capable of deep research that needs direct answers and raw content for analysis.

tools:
  tool_assigned:
    - name: Tavily
      config:
        API_key: ${TAVILY_API_KEY}
        max_results: 3
        include_answer: True
        include_raw_content: True
        search_depth: "advanced"

🚑 Troubleshooting​

  • API_key must be provided

    • The tool cannot find the API key. Ensure you have either passed it directly in the config or referenced a valid Key Vault variable (e.g., ${TAVILY_API_KEY}).

  • Other Errors

    • Check with the Tavily service or Rate Limits.
    • Check your billing status on the Tavily dashboard.
    • Ensure you aren't exceeding the max_results limit for your plan.