exa-cli

AI-powered web search, content extraction, and deep research from the command line via Exa API.

Built for speed. Optimized for AI agents — --compact mode strips all decoration for minimal token usage.

Install

cargo install exa-cli

From source

git clone https://github.com/Finesssee/exa-cli
cd exa-cli/rs
cargo build --release
# Binary at ./target/release/exa

Note: The Node.js version (npm install -g exa-cli) is deprecated. Use the Rust version above.

Setup

export EXA_API_KEY="your-key-here"   # Get one at https://exa.ai

Usage

# Search the web
exa search "latest rust async patterns" -n 5

# Compact mode (recommended for AI agents — auto-enabled on pipe)
exa search "query" --compact -n 3

# Only specific fields
exa search "query" --compact --fields url
exa search "query" --compact --fields title,url

# Tab-separated output
exa search "query" --tsv -n 5

# Find similar pages
exa find "https://example.com" --compact

# Extract page content
exa content https://example.com --compact

# Quick answer with sources
exa answer "what is WebAssembly" --compact

# Deep research
exa research "compare React vs Svelte in 2025" --compact

# Search types (instant is default — sub-150ms)
exa search "query" --type auto       # highest quality
exa search "query" --type fast       # balanced speed/quality
exa search "topic" --type deep       # comprehensive research

# Category filters
exa search "AI startups" --category company
exa search "Elon Musk" --category people

# Highlights (token-efficient excerpts)
exa search "react hooks" --highlights 3000

# Content freshness
exa search "breaking news" --max-age 1
exa search "historical data" --max-age -1   # cache only

# JSON output
exa search "query" --json --compact

Flags

Flag	Description
--compact	Terse output for AI/LLM consumption (auto on pipe)
--fields <list>	Comma-separated: title,url,date,content
--tsv	Tab-separated output (header + rows)
--max-chars <n>	Content truncation limit (default: 300 compact, 500 normal)
-n <num>	Number of results (default: 5)
--content	Include page text in search/find results
--highlights [n]	Key excerpts instead of full text (max chars, default: 2000)
--type <t>	Search type: instant (default, sub-150ms), auto, fast, deep, neural
--category <c>	Content category: company, people, tweet, news, research paper
--max-age <hrs>	Max content age in hours (0=always live, -1=cache only)
--verbosity <v>	Content verbosity: compact, standard, full
--json	JSON output (single-line with --compact)
--domain <d>	Restrict to domain
--after <date>	Published after YYYY-MM-DD
--before <date>	Published before YYYY-MM-DD
--no-cache	Bypass response cache
--cache-ttl <min>	Cache TTL in minutes (default: 60)
--no-sources	Hide sources in answer/research
--model <m>	exa-research (default) or exa-research-pro
--schema <file>	JSON schema for structured research output

Token Optimization

When used by AI agents, combine flags for minimal output:

# Absolute minimum — just URLs
exa search "query" --compact --fields url -n 3

# Titles + URLs only
exa search "query" --compact --fields title,url

# TSV for structured parsing
exa search "query" --tsv -n 5

# Cached repeat queries are instant (~13ms vs ~1200ms)
exa search "same query" --compact

License

MIT