exa-search_skill

Overview

This skill connects your app or CLI to Exa's AI-powered web search API to perform semantic search, extract page content, produce direct answers, and run long-running structured research. It packages endpoint choices and sensible defaults to help you fetch citations, summaries, and highlights for downstream AI workflows. Use it to turn web content into grounded, structured data for agents, reports, or RAG pipelines.

How this skill works

The skill reads an Exa API key from a local config file and sends JSON requests to the appropriate Exa endpoint (/search, /contents, /answer, /research/v1). /search performs semantic or keyword searches and can include extracted text, summaries, and highlights. /contents extracts full page text and highlights from provided URLs, /answer returns concise grounded responses, and /research/v1 runs asynchronous, schema-driven research tasks that produce structured output.

When to use it

• When you need semantic search results with extracted page text for a RAG system.
• When you want direct, citation-backed answers to factual questions.
• When extracting full content and highlights from a set of URLs.
• When you require multi-step, structured research reports or JSON outputs.
• When filtering results by date, domain, or content category for precise sourcing.

Best practices

• Store your API key in the local config (~/.config/exa-search/config.json) and validate JSON before requests.
• Pick endpoints by intent: search for discovery, contents for URL extraction, answer for quick facts, research for deep structured tasks.
• Limit search result counts to reduce cost and use date/domain filters to improve relevance.
• For research tasks provide a clear instructions and a small output schema (1–5 root fields) to improve reliability.
• Use livecrawl options only when freshness is required; prefer cache for lower cost and latency.

Example use cases

• Build a RAG pipeline: semantic search results with page text and summaries fed into an LLM.
• Automate competitive research: schedule research tasks that return structured comparisons of vendors.
• Create a news monitor: search with date filters and domain restrictions to surface recent coverage.
• Aggregate and summarize specific URLs: use /contents to pull and highlight key passages for reporting.
• Answer generation with citations: serve short, grounded answers to user queries with source links.