Exa

What it's good for

• 1

  Giving an agent live web search through a clean API instead of scraping

• 2

  Grounding a RAG pipeline in fresh, relevant sources

• 3

  Running deep-research queries that return many weighted sources programmatically

• 4

  Monitoring the web for changes on a topic and feeding updates into a system

• 5

  Pairing with an agent framework so the model can look things up mid-task

Pricing

• Free

  1,000 requests per month

  Free

• Search

  Per 1,000 requests

  $7/1k

• Deep Search

  Multi-source research queries

  $12/1k

• Agent modes

  Up to $2 per request by depth

  $0.025/req

How to use it

Grab an API key and start on the free 1,000 requests a month. Wire Exa into your agent or RAG pipeline as the retrieval step: send a query, get back ranked, clean results the model can read directly. Use Deep Search when one query should fan out to many sources, and Monitors when you need to react to web changes over time. Budget by request type, agent modes vary widely by depth, so meter the expensive ones. It pairs naturally with the agent tools in our directory.

Pros & cons

Frequently asked questions

• Is Exa free?

  The free tier covers 1,000 requests a month. Beyond that, pricing is per-request: Search $7 per 1,000, Deep Search $12 per 1,000, Answer $5, and Monitors $15, with agent modes from $0.025 to $2 per request depending on depth. Startup and education programs can get $1,000 in credits.

• Exa vs Perplexity: what is the difference?

  Perplexity is a human-facing chat-style answer UI. Exa is the opposite: search built for machines, exposing a clean API designed to feed agents and RAG pipelines rather than a results page for people. Pick Perplexity when a person wants the answer, Exa when an agent or system needs fresh, relevant sources programmatically.

• What is Exa best for?

  Giving an agent live web search through a clean API, grounding a RAG pipeline in fresh sources, running deep-research queries that return many weighted sources, and monitoring the web for changes to feed into a system. It pairs naturally with agent frameworks like LangGraph, CrewAI, and Lindy. For grounding only in your private docs, NotebookLM fits better.

• Can I use Exa with my agent or RAG pipeline?

  Yes, that is what it is built for. Grab an API key and wire Exa in as the retrieval step: send a query and get back ranked, clean results the model can read directly. Use Deep Search when one query should fan out to many sources, and Monitors when the system needs to react to web changes over time.

• How is Exa priced at scale?

  By request type, and the types vary widely in cost. Standard Search is $7 per 1,000 requests, Deep Search is $12, and agent modes range from $0.025 up to $2 per request by depth. Meter the expensive modes so a heavy agent run does not surprise you.

More

• Exa docs

Alternatives to Exa

Other tools we'd consider for the same job.

• Perplexity

  Search + synthesis — ask a question, get a sourced answer with citations you can actually click through and verify

• NotebookLM

  Upload sources, get a research assistant grounded in YOUR docs — with the Audio Overview podcast feature as a bonus

• Consensus

  AI search across 200M+ peer-reviewed papers. Get evidence-based answers, not hallucinated citations

Other Research tools

• Perplexity

  Search + synthesis — ask a question, get a sourced answer with citations you can actually click through and verify

• NotebookLM

  Upload sources, get a research assistant grounded in YOUR docs — with the Audio Overview podcast feature as a bonus

• Consensus

  AI search across 200M+ peer-reviewed papers. Get evidence-based answers, not hallucinated citations