Tavily Search

Get API Key: https://tavily.com

LiteLLM Python SDK

Tavily Search

import os
from litellm import search

os.environ["TAVILY_API_KEY"] = "tvly-..."

response = search(
    query="latest AI developments",
    search_provider="tavily",
    max_results=5
)

LiteLLM AI Gateway

1. Setup config.yaml

config.yaml

model_list:
  - model_name: gpt-6.1-sol
    litellm_params:
      model: gpt-6.1-sol
      api_key: os.environ/OPENAI_API_KEY

search_tools:
  - search_tool_name: tavily-search
    litellm_params:
      search_provider: tavily
      api_key: os.environ/TAVILY_API_KEY

2. Start the proxy

litellm --config /path/to/config.yaml

# RUNNING on http://0.0.0.0:4000

3. Test the search endpoint

Test Request

curl http://0.0.0.0:4000/v1/search/tavily-search \
  -H "Authorization: Bearer $LITELLM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "latest AI developments",
    "max_results": 5
  }'

Provider-specific Parameters

Tavily Search with Provider-specific Parameters

import os
from litellm import search

os.environ["TAVILY_API_KEY"] = "tvly-..."

response = search(
    query="latest tech news",
    search_provider="tavily",
    max_results=5,
    # Tavily-specific parameters
    topic="news",                    # 'general', 'news', 'finance'
    search_depth="advanced",         # 'basic', 'advanced'
    include_answer=True,             # Include AI-generated answer
    include_raw_content="markdown",  # Full page content, also accepts True or "text"
    include_images=True,
    include_favicon=True
)

Response

Tavily responses keep the standard title, url and snippet fields. snippet remains Tavily's content, while raw_content contains the full page content requested with include_raw_content. Result fields raw_content, score and favicon are included when Tavily returns them.

Tavily search response

{
  "object": "search",
  "results": [
    {
      "title": "Example article",
      "url": "https://example.com/article",
      "snippet": "A short excerpt from the article.",
      "date": null,
      "last_updated": null,
      "raw_content": "# Example article\n\nFull page content.",
      "score": 0.95,
      "favicon": "https://example.com/favicon.ico"
    }
  ],
  "provider_metadata": {
    "query": "latest tech news",
    "answer": "An answer based on the search results.",
    "images": ["https://example.com/image.png"],
    "request_id": "example-request-id",
    "response_time": 0.8
  }
}

provider_metadata preserves every top-level Tavily field except results. Depending on the request and Tavily's response, this includes answer, images, request_id, response_time and usage. Request include_answer, include_images or include_favicon to ask Tavily for those optional values. The same response mapping applies to the Python SDK and gateway search endpoints.

Cost

LiteLLM prices each Tavily search from the model cost map. A search sent with search_depth: "advanced" is billed at the tavily/search-advanced entry, which matches the 2 credits Tavily charges for it, and a search with any other depth, or none, is billed at tavily/search. This applies to the SDK, every /search route on the proxy, and web searches the proxy runs for web search interception.

To bill your own rate for every depth, set input_cost_per_query on the search tool, as described in Custom Per-Query Pricing. A price you register for tavily/search with litellm.register_model keeps applying to advanced searches until you also register tavily/search-advanced.