exa

name	exa
description	Use this skill when the user asks for "exa", "exa search", "deep research", "web search with citations", "answer with sources", "find similar websites", "find similar sites", or needs AI-powered web search, semantic search, or agentic research over the live web. Wraps the Exa AI API (search, contents, answer, findSimilar, research).

Exa - AI Web Search and Research

Exa is a search API built for agents. Use it when the user wants:

• High-signal web results with full-text content or highlights (exa search)
• A cited answer to a specific question (exa answer)
• Pages similar to a known URL (exa similar)
• A full multi-step research report on a topic (exa research)
• Raw text/summary of specific URLs (exa contents)

Setup

See SETUP.md. Key goes in ~/.exa/config.json (or EXA_API_KEY env var).

Commands

All commands print JSON. Pipe to jq for filtering.

Search

exa search "latest research on reasoning models" --num 10
exa search "diffusion music generation papers" --type neural --category "research paper"
exa search "founder blogs about RLHF" --text
exa search "ai agents startups 2026" --highlights
exa search "foo" --start-published 2026-01-01 --end-published 2026-04-24

Key flags: --num N (results, default 10), --type {auto,fast,neural,keyword}, --category {company,research paper,news,pdf,github,tweet,personal site,linkedin profile,financial report}, --text, --highlights, --summary, --include-domain, --exclude-domain, --start-published YYYY-MM-DD, --end-published YYYY-MM-DD.

Answer (with citations)

exa answer "What is the latest valuation of SpaceX?"
exa answer "What did Anthropic announce this week?" --text

Returns {answer, citations[], costDollars}. Each citation includes URL, title, author, publishedDate, optional text.

Find similar

exa similar "https://arxiv.org/abs/2307.06435" --num 10
exa similar "https://example.com/blog" --exclude-source-domain

Contents (fetch URL text/summary)

exa contents https://arxiv.org/abs/2307.06435 --text
exa contents https://example.com/a https://example.com/b --summary "key claims"
exa contents https://long.page --text --max-chars 4000

Deep research

exa research "state of small language models in 2026" --model exa-research
exa research "compare vector DBs for agent memory" --model exa-research-pro --wait

Creates an async research task. Without --wait, returns the task ID so you can poll later with exa research status <id>. With --wait, polls until complete (up to ~10 min) and prints the final report.

Models:

• exa-research-fast - quick scan, cheapest
• exa-research - default, balanced
• exa-research-pro - deepest, slowest, most expensive

Typical Usage Patterns

• Literature review: exa search "<topic>" --category "research paper" --num 20 --text | jq '.results[] | {title, url, publishedDate}'
• Answer with sources: exa answer "<question>" then cite .citations[].url in your reply
• Find alternatives: exa similar "<url>" to discover competitors/similar content
• Briefing: exa research "<topic>" --wait for a full multi-source synthesis

Notes

• Every response includes a costDollars field. Mention cost to the user for research calls (they can add up).
• --text significantly increases token count. Use --highlights or --summary when you just need gist.
• Exa ranks by semantic relevance. For exact-match queries (names, SKUs), pass --type keyword.
• Installed via uv tool install --editable ~/agent/skills/exa/cli.