Exa Web Search

Search the web from an agent with filters for type, date range, domains, and result depth

How Exa Web Search Works

Use Exa Web Search when your agent needs current external information that is not available in Flywheel event or profile data. The tool executes a real-time web query through the Exa search API and returns structured results the agent can reason over to personalize messages, verify claims, or gather competitive context. You control scope with domain allowlists/blocklists, date ranges, and content category filters so the agent only surfaces relevant, approved sources.

Configuration

Inputs

Prompt · string

Instructions for when the agent should run web search and how it should use retrieved sources.

Query · string · required

Search query text. Can be user-defined or AI-defined.

Maximum Results · number · default: "10"

Number of results to return.

Search Type · enum · default: "auto"

Search strategy. Options: auto, fast, deep, neural.

Content Category · enum

Optional source category filter. Options: company, research paper, news, pdf, tweet, personal site, financial report, people.

Date Range · object

Relative date window using value + unit (days, weeks, months, years).

Include Domains · string[]

Domain allowlist.

Exclude Domains · string[]

Domain blocklist.

Include Text / Exclude Text · string[]

Optional text constraints for filtering results.

User Location · string

ISO country code context (for localized result relevance).

Moderation · boolean

Content moderation flag.

Use Autoprompt · boolean

Allow Exa autoprompt behavior.

Outputs

results · object[] · required

Array of search result objects returned by the Exa API for the agent to reason over.

results[].url · string · required

The URL of the matching web page.

results[].title · string · required

The page title as indexed by the search provider.

results[].snippet · string

A text excerpt from the page content relevant to the query.

result_count · number · required

Total number of results returned in this search execution.

query_used · string · required

The final query string sent to Exa, which may differ from the input if autoprompt rewrote it.

Use Cases

Competitive Intelligence

Query: "{{user.competitor}} product launch 2025"
Content Category: news
Date Range: 30 days
Use: Agent references recent competitor moves to personalize renewal or upsell messaging

Account Enrichment

Query: "{{user.company}} funding round OR acquisition"
Content Category: news
Include Domains: techcrunch.com, crunchbase.com
Use: Agent enriches account context with latest funding or M&A activity before outreach

Content Research

Query: "best practices {{user.industry}} {{event.topic}}"
Content Category: research paper
Maximum Results: 5
Use: Agent gathers authoritative sources to include in educational drip campaigns

Compliance Verification

Query: "{{user.company}} regulatory action OR compliance update"
Include Domains: sec.gov, finra.org
Exclude Domains: reddit.com, twitter.com
Use: Agent validates compliance standing before sending sensitive communications

Best Practices

Query Strategy

• Keep queries specific and task-oriented to reduce noise in results.
• Use dynamic template variables (e.g., {{user.company}}) so queries stay relevant per run.
• Prefer the auto search type unless you have a clear reason to force deep or neural.
• Limit result counts to what the agent actually needs — fewer results mean faster execution.

Domain Control

• Constrain domains for compliance-sensitive messaging to approved sources only.
• Use the exclude list to block social media or user-generated content when accuracy matters.
• Combine domain filters with content category filters for the tightest result scope.
• Regularly review your allowlists to ensure referenced domains are still authoritative and active.

Performance

• Use lower result counts for speed; increase only when the agent needs breadth.
• Set date ranges to avoid stale content and keep results timely.
• Enable autoprompt only when the agent’s query may benefit from Exa’s reformulation logic.
• Monitor search usage to stay within rate limits and avoid unnecessary API calls.