exa-cli - SKILL.md Agent Skill

name: exa-cli description: CLI tool for Exa AI search API. Search the web semantically, extract content from URLs, get AI answers with citations, and conduct automated research. Use when the user wants to search the web, find information, extract content from websites, get AI answers with sources, or perform automated research tasks.

Exa CLI Skill

This skill provides command-line access to Exa AI search API for semantic web search, content extraction, AI-powered Q&A, and automated research.

Capabilities

Semantic Web Search: Search by meaning, not just keywords
Content Extraction: Pull full text, highlights, and summaries from URLs
AI-Powered Answers: Get answers with source citations
Similar Page Discovery: Find pages similar to a given URL
Automated Research: Deep research tasks with polling and status tracking

Setup

1. Installation

# Option A: Install globally
npm install -g @ai-bubble/exa-cli

# Option B: Use with npx (no installation)
npx @ai-bubble/exa-cli <command>

# Option C: Install with Bun
bun install @ai-bubble/exa-cli

2. Configuration

Set the Exa API key:

export EXA_API_KEY="your-api-key-here"

Or pass with each command:

exa search "query" --api-key "your-key"

Get an API key at: https://dashboard.exa.ai/api-keys

Core Skills

Skill: Semantic Search

Search the web using semantic understanding.

When to use:

User asks to search for information
User wants to find articles, papers, or websites
User needs to research a topic

Command:

exa search "<query>" [options]

Key Options:

--num-results <n>: Results count (default: 10, max: 10 on basic plans)
--type <mode>: Search mode - auto, fast, deep, instant
--text: Include full text content
--highlights: Include relevant text highlights
--summary: Include AI-generated summary
--category <type>: Filter by category (company, research paper, news, pdf, tweet, personal site, financial report, people)
--include-domains <list>: Comma-separated domains to include
--exclude-domains <list>: Comma-separated domains to exclude
--start-date <date>: Start date (YYYY-MM-DD)
--end-date <date>: End date (YYYY-MM-DD)
--autoprompt: Enhance query with autoprompt

Examples:

# Basic semantic search
exa search "latest AI developments"

# Deep search with content extraction
exa search "transformer architecture explained" --type deep --text --highlights

# Filter by category and date
exa search "startup funding" --category news --start-date 2024-01-01

# Domain-specific search
exa search "machine learning" --include-domains "arxiv.org,openai.com"

Skill: Content Extraction

Retrieve content from specific URLs.

When to use:

User provides a URL and wants to extract content
User wants to analyze a specific article or page
User needs text from academic papers or documentation

Command:

exa contents <url1> [<url2> ...] [options]

Key Options:

--text: Extract full text content
--highlights: Extract relevant highlights
--summary: Generate AI summary
--max-age-hours <n>: Maximum cached content age (0 = always fetch fresh)

Examples:

# Extract text from single URL
exa contents "https://arxiv.org/abs/2304.15004" --text

# Extract from multiple URLs
exa contents "https://example.com/1" "https://example.com/2" --text --highlights

# Force fresh content fetch
exa contents "https://example.com/article" --text --max-age-hours 0

Skill: Find Similar Pages

Discover pages similar to a given URL.

When to use:

User wants to find similar articles or resources
User wants related content to a specific page
User is researching a topic and found one good source

Command:

exa similar <url> [options]

Key Options:

--num-results <n>: Number of similar pages (default: 10)
--exclude-source-domain: Don't return pages from same domain
--text, --highlights, --summary: Include content
--category <type>: Filter by category

Examples:

# Find similar pages
exa similar "https://openai.com/research/gpt-4"

# Exclude same domain, get content
exa similar "https://techcrunch.com/article" --exclude-source-domain --text

Skill: AI-Powered Answers

Get AI-generated answers with source citations.

When to use:

User asks a factual question
User wants an explanation with sources
User needs a quick answer with references

Command:

exa answer "<question>" [options]

Key Options:

--text: Include source text in citations
--model <model>: Model to use (exa, exa-pro)
--stream: Stream response in real-time
--system-prompt <prompt>: Custom system prompt

Examples:

# Basic question
exa answer "What is quantum computing?"

# Streaming response
exa answer "Explain neural networks" --stream

# Pro model for complex queries
exa answer "Compare transformer vs RNN architectures" --model exa-pro

# Custom explanation style
exa answer "Explain blockchain" --system-prompt "Explain to a 10-year-old"

Skill: Automated Research

Create and manage deep research tasks.

When to use:

User needs comprehensive research on a topic
User wants automated information gathering
User needs structured research output

Command:

exa research "<instructions>" [options]

Key Options:

--model <model>: Research model (fast, regular, pro)
--poll: Wait for completion and show results
--poll-interval <ms>: Check interval in ms (default: 1000)
--timeout <ms>: Max wait time (default: 600000 = 10 min)

Examples:

# Create research task
exa research "Latest SpaceX valuation and funding rounds"

# Create and wait for results
exa research "CRISPR applications in medicine" --poll

# Pro model with custom polling
exa research "Fusion energy progress 2024" --model pro --poll --timeout 300000

Skill: Research Task Management

Check status and list research tasks.

When to use:

User wants to check on a research task
User needs to list all their research tasks
User wants to see research results

Commands:

# Check specific task
exa research-status <task-id>

# List all tasks
exa research-list

# List with pagination
exa research-list --limit 10 --cursor "next-page-token"

Workflows

Workflow: Deep Topic Research

Research a topic thoroughly with multiple approaches:

Initial Search

exa search "topic keywords" --type deep --num-results 10 --text

Extract Key Sources

exa contents "https://key-source-1.com" "https://key-source-2.com" --text --highlights

Find Related Content

exa similar "https://best-article-found.com" --num-results 5 --text

Get AI Summary

exa answer "Synthesize findings about [topic]" --model exa-pro

Deep Research (if needed)

exa research "Comprehensive analysis of [topic]" --model pro --poll

Workflow: News Monitoring

Track recent news on a topic:

Search Recent News

exa search "topic" --category news --start-date $(date -d '7 days ago' +%Y-%m-%d) --num-results 10

Get Full Articles

exa contents "https://news-site.com/article-1" "https://news-site.com/article-2" --text --summary

Workflow: Academic Research

Research academic papers:

Search Papers

exa search "research topic" --category "research paper" --num-results 10

Extract Paper Content

exa contents "https://arxiv.org/abs/xxxxx" --text --highlights

Find Related Papers

exa similar "https://arxiv.org/abs/xxxxx" --category "research paper"

Common Patterns

Pattern: JSON Output for Scripting

Use --json flag when integrating with scripts:

# Get results as JSON
exa search "query" --json | jq '.results[0].title'

# Check task status programmatically
exa research-status "task-id" --json | jq '.status'

Pattern: Content Options

Content flags (--text, --highlights, --summary) work across multiple commands:

# Search with content
exa search "query" --text --highlights

# Similar with content
exa similar "url" --text --summary

# Contents extraction
exa contents "url" --text --highlights --summary

Pattern: Date Filtering

Filter by publication date:

# Last 30 days
exa search "topic" --start-date $(date -d '30 days ago' +%Y-%m-%d)

# Specific date range
exa search "topic" --start-date 2024-01-01 --end-date 2024-12-31

Pattern: Domain Control

Include or exclude specific domains:

# Search only specific domains
exa search "topic" --include-domains "arxiv.org,openai.com,anthropic.com"

# Exclude domains
exa search "topic" --exclude-domains "reddit.com,twitter.com"

Search Categories

Available categories for filtering:

company - Company websites and profiles
research paper - Academic papers and research
news - News articles and journalism
pdf - PDF documents
tweet - Twitter/X posts
personal site - Personal blogs and websites
financial report - Financial documents
people - People profiles

Search Types

Different search modes for different needs:

auto - Automatically choose best type (default)
fast - Quick search, good for speed-critical apps
deep - Thorough search, best for research
instant - Lowest latency, optimized for real-time

Research Models

Research task quality levels:

fast / exa-research-fast - Fastest, good for quick facts
regular / exa-research - Balanced speed and quality
pro / exa-research-pro - Highest quality, slower

Error Handling

Common errors and solutions:

Missing API Key:

Error: EXA_API_KEY not set. Use --api-key or set EXA_API_KEY environment variable.

Solution: Set EXA_API_KEY environment variable or use --api-key flag

Missing Arguments:

Error: search requires a query argument

Solution: Provide the required argument

Invalid URLs:

Error: API Error 400: Invalid URL format

Solution: Check URL format and try again

Rate Limiting:

Error: API Error 429: Too many requests

Solution: Wait and retry, or upgrade your Exa plan

Tips and Best Practices

Use autoprompt for better results: Add --autoprompt to searches when you're not getting relevant results
Combine content options: Use --text --highlights --summary together for comprehensive content extraction
Stream for long answers: Use --stream with the answer command for better UX on long responses
Poll for research: Always use --poll with research tasks unless you want to check status manually later
Check costs: Look for cost information in the output to monitor API usage
Use categories: Filter by category (especially research paper or news) for more targeted results
Date filtering: Use date filters for time-sensitive research
JSON for automation: Use --json flag when integrating with other tools or scripts

Resources

Full Documentation: https://github.com/sandiiarov/exa-cli#readme
Exa API Docs: https://exa.ai/docs
Issue Tracker: https://github.com/sandiiarov/exa-cli/issues
npm Package: https://www.npmjs.com/package/@ai-bubble/exa-cli

Examples Summary

# Search
exa search "AI startups" --num-results 5 --text --type deep

# Contents
exa contents "https://example.com/article" --text --highlights

# Similar
exa similar "https://openai.com/blog" --exclude-source-domain

# Answer
exa answer "What is LLM?" --model exa-pro --stream

# Research
exa research "SpaceX history" --model pro --poll
exa research-status "task-id"
exa research-list --limit 10