Exa Search Plugin

#Installation

Install with the Exa extra:

bash

pip install 'daita-agents[exa]'

Or install the SDK dependency directly:

bash

pip install exa-py

#Quick Start

python

import os
 
from daita import Agent
from daita.plugins import exa_search
 
# Uses EXA_API_KEY from the environment
search = exa_search()
 
agent = Agent(
    name="Research Assistant",
    prompt="You are a research assistant. Use search when current or sourced information matters.",
    tools=[search],
    model="gpt-4o-mini",
)
 
await agent.start()
result = await agent.run("Find recent research on agent evaluation systems.")
await agent.stop()

#Getting Started with Exa

Create an Exa account at exa.ai
Get an API key from the dashboard
Set the environment variable:

bash

export EXA_API_KEY=your-key

Create the plugin with exa_search()

#Direct Usage

Use the plugin directly when you need programmatic search without an agent:

python

from daita.plugins import exa_search
 
async with exa_search(num_results=5, search_type="auto") as search:
    results = await search.search(
        "Python async best practices",
        include_highlights=True,
        include_summary=True,
    )
 
    for result in results["results"]:
        print(result["title"])
        print(result["url"])
        print(result["snippet"])

#Connection Parameters

python

exa_search(
    api_key: Optional[str] = None,
    num_results: int = 5,
    search_type: str = "auto",
    include_text: bool = False,
    include_highlights: bool = True,
    include_summary: bool = False,
    max_text_chars: int = 1000,
    category: Optional[str] = None,
    include_domains: Optional[List[str]] = None,
    exclude_domains: Optional[List[str]] = None,
    **kwargs,
)

#Parameters

api_key (str): Exa API key. Defaults to EXA_API_KEY.
num_results (int): Default number of results to return. Defaults to 5.
search_type (str): Search mode. One of "auto", "neural", "fast", or "keyword".
include_text (bool): Include full page text in each result. Defaults to False.
include_highlights (bool): Include Exa-generated highlights. Defaults to True.
include_summary (bool): Include Exa-generated summaries. Defaults to False.
max_text_chars (int): Maximum text characters per result when include_text=True.
category (str): Optional default category filter.
include_domains (List[str]): Optional default domain allowlist.
exclude_domains (List[str]): Optional default domain blocklist.

#Search

python

async with exa_search() as search:
    results = await search.search(
        query="latest vector database benchmarks",
        num_results=10,
        search_type="neural",
        category="research paper",
        include_domains=["arxiv.org"],
        start_published_date="2025-01-01",
        include_highlights=True,
        include_summary=True,
    )

search() returns:

python

{
    "success": True,
    "query": "latest vector database benchmarks",
    "results": [
        {
            "title": "...",
            "url": "...",
            "snippet": "...",
            "score": 0.91,
            "published_date": "...",
            "author": "...",
            "text": "...",
            "highlights": ["..."],
            "summary": "...",
        }
    ],
    "count": 1,
}

#Find Similar Pages

Use find_similar() when you have one useful source and want related pages:

python

async with exa_search(num_results=5) as search:
    related = await search.find_similar(
        "https://example.com/reference-article",
        include_highlights=True,
        include_summary=True,
    )

#Using with Agents

Exa exposes search operations as tools that agents can call autonomously:

python

import os
 
from daita import Agent
from daita.plugins import exa_search
 
search = exa_search(
    api_key=os.getenv("EXA_API_KEY"),
    num_results=6,
    search_type="auto",
    include_highlights=True,
    include_summary=True,
)
 
agent = Agent(
    name="Market Researcher",
    prompt="""You are a careful research agent.
Use search_web for current information and find_similar when a source looks useful.
Prefer authoritative sources, compare multiple results, and include URLs in the answer.
""",
    tools=[search],
    model="gpt-4o-mini",
)
 
await agent.start()
result = await agent.run("Research recent adoption trends for lakehouse architectures.")
await agent.stop()

#Available Tools

Tool	Description	Parameters
search_web	Semantic web search with optional content fields	`query` (required), `num_results`, `search_type`, `category`, `include_domains`, `exclude_domains`, `start_published_date`, `end_published_date`
find_similar	Find pages semantically similar to a URL	`url` (required), `num_results`, `include_text`, `include_highlights`, `include_summary`

Tool Category: search Tool Source: plugin Plugin Name: ExaSearch

#Search Types

Type	Use when
`auto`	You want Exa to choose the best search mode for the query.
`neural`	You want semantic ranking for conceptual or research-heavy questions.
`fast`	You want lower-latency search for straightforward lookups.
`keyword`	You need lexical matching for exact terms, names, identifiers, or codes.

#Category Filters

Supported categories:

company
research paper
news
personal site
financial report
people

#Domain and Date Filters

python

async with exa_search() as search:
    results = await search.search(
        "OpenAI structured outputs",
        include_domains=["openai.com"],
        exclude_domains=["community.example.com"],
        start_published_date="2025-01-01",
        end_published_date="2026-01-01",
    )

Domain filters are useful when an agent should stay close to primary sources or avoid low-signal sites. Date filters are useful for news, market research, and rapidly changing technical topics.

#Error Handling

The plugin maps Exa/API failures into Daita's error hierarchy:

Condition	Error type
Rate limit or HTTP 429	`RateLimitError`
Invalid or missing API key	`AuthenticationError`
Forbidden or bad request	`PermanentError`
Request timeout	`TimeoutError`
Network or connection issue	`ConnectionError`
Temporary service issue	`TransientError`
Other search execution error	`RetryableError`

#Best Practices

Use EXA_API_KEY instead of hardcoding secrets.
Keep include_text=False unless the agent needs full page content.
Use include_highlights=True for concise source evidence.
Use domain filters for trusted-source research.
Use find_similar() to expand from a high-quality source.
Prefer search_type="auto" unless you know the query needs semantic or exact matching.

WebSearch Plugin - Tavily-backed web search, news search, and URL fetching
Agent - Adding plugins to agents
Error Handling - Daita exception types and retry behavior

Exa Search Plugin

#Installation

#Quick Start

#Getting Started with Exa

#Direct Usage

#Connection Parameters

#Parameters

#Search

#Find Similar Pages

#Using with Agents

#Available Tools

#Search Types

#Category Filters

#Domain and Date Filters

#Error Handling

#Best Practices

#Related