Exa Search

Exa Search

Exa is a search API built for AI. Five endpoints, each for a different job.

Setup

# API key from pass or env
export EXA_API_KEY=$(pass api/exa 2>/dev/null)
# Or: export EXA_API_KEY=your-key
# Dashboard & credits: https://dashboard.exa.ai

Endpoints

1. Search

curl -s -X POST "https://api.exa.ai/search" \
  -H "x-api-key: $EXA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "query": "transformer architecture explained",
    "type": "auto",
    "numResults": 10,
    "contents": { "text": { "maxCharacters": 3000 } }
  }' | jq '.results[] | {title, url, text}'

Search types

Categories (optional filter)

company, research paper, news, pdf, tweet, personal site, financial report, people

Filters

{
  "query": "climate policy 2025",
  "numResults": 20,
  "category": "research paper",
  "includeDomains": ["nature.com", "science.org"],
  "excludeDomains": ["wikipedia.org"],
  "startPublishedDate": "2025-01-01",
  "endPublishedDate": "2025-12-31",
  "includeText": ["carbon tax"],
  "excludeText": ["opinion"],
  "contents": {
    "text": { "maxCharacters": 2000 },
    "highlights": true,
    "summary": true
  }
}

Content options

Request alongside search results:

Livecrawl: "livecrawl": "always" fetches fresh content instead of cache.

2. Deep Search

Same /search endpoint with type: "deep". Returns higher quality results through multi-query expansion and reasoning.

curl -s -X POST "https://api.exa.ai/search" \
  -H "x-api-key: $EXA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "query": "What are the latest battery breakthroughs?",
    "type": "deep",
    "numResults": 10,
    "answer": true,
    "contents": { "text": true }
  }' | jq '{answer, results: [.results[] | {title, url}]}'

Deep search with structured output

curl -s -X POST "https://api.exa.ai/search" \
  -H "x-api-key: $EXA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "query": "Compare React vs Vue in 2025",
    "type": "deep",
    "answer": true,
    "effort": "max",
    "outputSchema": {
      "type": "object",
      "properties": {
        "summary": { "type": "string" },
        "react_pros": { "type": "array", "items": { "type": "string" } },
        "vue_pros": { "type": "array", "items": { "type": "string" } },
        "recommendation": { "type": "string" }
      },
      "required": ["summary"]
    },
    "contents": { "text": true }
  }' | jq '.'

Deep search effort levels

Additional queries

Provide alternative phrasings to broaden deep search coverage:

{
  "query": "autonomous vehicle safety statistics",
  "type": "deep",
  "additionalQueries": [
    "self-driving car accident rates 2024-2025",
    "waymo cruise safety data comparison",
    "NHTSA autonomous vehicle crash reports"
  ]
}

3. Find Similar

Given a URL, find semantically similar pages.

curl -s -X POST "https://api.exa.ai/findSimilar" \
  -H "x-api-key: $EXA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "url": "https://paulgraham.com/greatwork.html",
    "numResults": 10,
    "excludeSourceDomain": true,
    "contents": { "text": { "maxCharacters": 1000 } }
  }' | jq '.results[] | {title, url}'

Supports same filters as search: includeDomains, excludeDomains, date ranges, category.

4. Contents

Extract content from known URLs (no search needed).

curl -s -X POST "https://api.exa.ai/contents" \
  -H "x-api-key: $EXA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "urls": ["https://openai.com/research"],
    "text": true,
    "summary": true,
    "highlights": { "query": "safety research" }
  }' | jq '.'

Options: subpages (crawl N subpages), subpageTarget (specific paths), extras (links, imageLinks).

5. Answer

Direct answer to a question, backed by search results.

curl -s -X POST "https://api.exa.ai/answer" \
  -H "x-api-key: $EXA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "query": "What caused the 2008 financial crisis?",
    "numResults": 5
  }' | jq '{answer, citations: [.results[] | {title, url}]}'

Streaming:

curl -s -X POST "https://api.exa.ai/answer" \
  -H "x-api-key: $EXA_API_KEY" \
  -H "content-type: application/json" \
  -d '{"query": "Explain quantum computing", "stream": true}'

6. Research (Async Deep Research)

Agent-style research: Exa plans, searches, reads, reasons, and produces a structured report. Runs async — create a task, poll for results.

Models

Create a research task

curl -s -X POST "https://api.exa.ai/research" \
  -H "x-api-key: $EXA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "instructions": "What are the leading open-source LLMs in 2025? Compare their benchmarks, licensing, and real-world adoption.",
    "model": "exa-research",
    "outputSchema": {
      "type": "object",
      "properties": {
        "models": {
          "type": "array",
          "items": {
            "type": "object",
            "properties": {
              "name": { "type": "string" },
              "organization": { "type": "string" },
              "license": { "type": "string" },
              "benchmark_score": { "type": "string" },
              "adoption_notes": { "type": "string" }
            }
          }
        },
        "summary": { "type": "string" }
      },
      "required": ["models", "summary"]
    }
  }' | jq '{researchId: .researchId, status: .status}'

Poll for results

RESEARCH_ID="your-research-id"
curl -s "https://api.exa.ai/research/$RESEARCH_ID" \
  -H "x-api-key: $EXA_API_KEY" | jq '{status, output}'

Stream events (SSE)

curl -s "https://api.exa.ai/research/$RESEARCH_ID?stream=true" \
  -H "x-api-key: $EXA_API_KEY" \
  -H "accept: text/event-stream"

Events: research-definition → plan-definition → plan-operation (think/search/crawl) → task-* → research-output

List research tasks

curl -s "https://api.exa.ai/research?limit=10" \
  -H "x-api-key: $EXA_API_KEY" | jq '.data[] | {researchId, status, instructions}'

Python SDK (recommended for research)

uv pip install exa-py

from exa_py import Exa
import os, time

exa = Exa(os.environ["EXA_API_KEY"])

# Create task
task = exa.research.create(
    instructions="Summarize recent advances in fusion energy",
    model="exa-research",
    output_schema={
        "type": "object",
        "properties": {
            "summary": {"type": "string"},
            "key_developments": {"type": "array", "items": {"type": "string"}}
        }
    }
)

# Poll until done
result = exa.research.poll_until_finished(
    task.research_id,
    poll_interval=2000,
    timeout_ms=300000
)

if result.status == "completed":
    print(result.output.content)
    print(f"Cost: ${result.cost_dollars.total:.4f}")

Async version:

from exa_py import AsyncExa
exa = AsyncExa(os.environ["EXA_API_KEY"])
task = await exa.research.create(instructions="...", model="exa-research-fast")
result = await exa.research.poll_until_finished(task.research_id)

Batch Queries Pattern

Run many searches, deduplicate by URL:

QUERIES=(
  "topic A primary sources"
  "topic A academic research"
  "topic A recent news 2025"
)

for q in "${QUERIES[@]}"; do
  curl -s -X POST "https://api.exa.ai/search" \
    -H "x-api-key: $EXA_API_KEY" \
    -H "content-type: application/json" \
    -d "{
      \"query\": \"$q\",
      \"numResults\": 8,
      \"excludeDomains\": [\"wikipedia.org\"],
      \"contents\": { \"text\": { \"maxCharacters\": 2000 } }
    }" | jq -c '.results[]' >> /tmp/exa-batch.jsonl
  sleep 0.5
done

jq -s 'unique_by(.url)' /tmp/exa-batch.jsonl > /tmp/exa-deduped.json
echo "Unique: $(jq length /tmp/exa-deduped.json)"

Costs

Check credits: https://dashboard.exa.ai

Response Shape

All search endpoints return:

{
  "requestId": "...",
  "results": [
    {
      "title": "...",
      "url": "...",
      "publishedDate": "...",
      "score": 0.95,
      "text": "...",
      "highlights": ["..."],
      "summary": "..."
    }
  ],
  "costDollars": { "total": 0.003 },
  "searchType": "neural",
  "answer": "..." // only with answer=true or /answer endpoint
}