Exa

ExaTools enable an Agent to search the web using Exa, retrieve content from URLs, find similar content, and get AI-powered answers.

Prerequisites

The following examples require the exa-py and openai libraries and an API key which can be obtained from Exa.

uv pip install -U agno exa-py openai

export EXA_API_KEY=***

Example

The following agent will search Exa for AAPL news and print the response.

cookbook/91_tools/exa_tools.py

from agno.agent import Agent
from agno.tools.exa import ExaTools

agent = Agent(
    tools=[ExaTools(
        include_domains=["cnbc.com", "reuters.com", "bloomberg.com"],
        category="news",
        show_results=True,
        text_length_limit=1000,
    )],
    )
agent.print_response("Search for AAPL news", markdown=True)

Toolkit Functions

Function	Description
`search_exa`	Searches Exa for a query with optional category filtering
`get_contents`	Retrieves detailed content from specific URLs
`find_similar`	Finds similar content to a given URL
`exa_answer`	Gets an AI-powered answer to a question using Exa search results
`research`	Performs deep research on a topic and returns structured results with citations

Toolkit Params

Parameter	Type	Default	Description
`enable_search`	`bool`	`True`	Enable search functionality
`enable_get_contents`	`bool`	`True`	Enable content retrieval
`enable_find_similar`	`bool`	`True`	Enable finding similar content
`enable_answer`	`bool`	`True`	Enable AI-powered answers
`enable_research`	`bool`	`False`	Enable research functionality
`all`	`bool`	`False`	Enable all functionality
`api_key`	`Optional[str]`	`None`	Exa API key. Uses `EXA_API_KEY` environment variable if not set
`text`	`bool`	`True`	Include text content in results
`text_length_limit`	`int`	`1000`	Maximum length of text content per result
`summary`	`bool`	`False`	Requests provider summaries, but the current result serializer omits them
`num_results`	`Optional[int]`	`None`	Default number of results
`livecrawl`	`str`	`"always"`	Stored but not forwarded to the current provider call
`start_crawl_date`	`Optional[str]`	`None`	Include results crawled after date (YYYY-MM-DD)
`end_crawl_date`	`Optional[str]`	`None`	Include results crawled before date (YYYY-MM-DD)
`start_published_date`	`Optional[str]`	`None`	Include results published after date (YYYY-MM-DD)
`end_published_date`	`Optional[str]`	`None`	Include results published before date (YYYY-MM-DD)
`type`	`Optional[str]`	`None`	Search strategy (for example `auto`, `fast`, or `deep`), not a content format; use `category` for supported categories
`category`	`Optional[str]`	`None`	Category filter (e.g., news, research paper)
`include_domains`	`Optional[List[str]]`	`None`	Restrict results to these domains
`exclude_domains`	`Optional[List[str]]`	`None`	Exclude results from these domains
`show_results`	`bool`	`False`	Log search results for debugging
`model`	`Optional[str]`	`None`	Answer model used by `exa_answer` ('exa' or 'exa-pro'), not search strategy
`timeout`	`int`	`30`	Worker-result timeout; not a strict wall-clock bound and not applied to research polling
`research_model`	`Literal["exa-research", "exa-research-pro"]`	`"exa-research"`	Model used by the `research` function

Current response and timeout boundaries

livecrawl is not forwarded by this adapter, and requested summaries are dropped from its formatted results. Use the returned text excerpts, or the provider SDK directly when you need those fields.

The timeout wrapper waits for its worker during cleanup, so a slow underlying request can outlast timeout. Research polling follows a separate path. Use provider-level request timeouts in an application that needs a strict deadline. See Exa search parameters for current strategy values.

Developer Resources

Tools