Exa Performance Tuning

---
name: exa-performance-tuning
description: |
  Optimize Exa API performance with caching, batching, and connection pooling.
  Use when experiencing slow API responses, implementing caching strategies,
  or optimizing request throughput for Exa integrations.
  Trigger with phrases like "exa performance", "optimize exa",
  "exa latency", "exa caching", "exa slow", "exa batch".
allowed-tools: Read, Write, Edit
version: 1.0.0
license: MIT
author: Jeremy Longshore <jeremy@intentsolutions.io>
compatible-with: claude-code, codex, openclaw
---
# Exa Performance Tuning

## Overview
Optimize Exa AI search API response times and throughput for production RAG pipelines and search integrations. Exa search latency varies by type: keyword search (200-500ms), neural search (500-2000ms), and auto mode (300-1500ms).

## Prerequisites
- Exa API integration (`exa-js` SDK or REST API)
- Cache infrastructure (Redis or in-memory LRU)
- Understanding of search patterns in your application

## Instructions

### Step 1: Choose Search Type by Latency Requirement
```typescript
import Exa from 'exa-js';

// Match search type to latency budget
function optimizedSearch(exa: Exa, query: string, latencyBudgetMs: number) {
  if (latencyBudgetMs < 500) {  # HTTP 500 Internal Server Error
    // Fast path: keyword search for structured/exact queries
    return exa.search(query, { type: 'keyword', numResults: 3 });
  } else if (latencyBudgetMs < 1500) {  # 1500 = configured value
    // Balanced: auto mode picks best approach
    return exa.search(query, { type: 'auto', numResults: 5 });
  } else {
    // Quality: neural search for semantic understanding
    return exa.search(query, { type: 'neural', numResults: 10 });
  }
}
```

### Step 2: Cache Search Results
```typescript
import { LRUCache } from 'lru-cache';

const searchCache = new LRUCache<string, any>({
  max: 10000,  # 10000: 10 seconds in ms
  ttl: 2 * 3600_000, // 2-hour TTL for most searches
});

async function cachedSearch(exa: Exa, query: string, options: any) {
  const key = `${query}:${options.type}:${options.numResults}`;
  const cached = searchCache.get(key);
  if (cached) return cached; // Cache hit: 0ms vs 500-2000ms  # HTTP 500 Internal Server Error

  const results = await exa.search(query, options);
  searchCache.set(key, results);
  return results;
}
```

### Step 3: Minimize Result Count
```typescript
// Each additional result adds latency (content retrieval)
const RESULT_CONFIGS: Record<string, number> = {
  'rag-context':     3,   // Only need top 3 for RAG
  'autocomplete':    5,   // Quick suggestions
  'deep-research':  10,   // Comprehensive coverage
};

// Don't default to numResults: 10 when 3 suffices
// Reducing from 10 to 3 results saves ~200-500ms per search  # HTTP 200 OK
```

### Step 4: Parallelize Independent Searches
```typescript
// When RAG needs multiple search contexts, run them in parallel
async function parallelContextSearch(exa: Exa, queries: string[]) {
  const searches = queries.map(q =>
    cachedSearch(exa, q, { type: 'auto', numResults: 3 })
  );
  return Promise.all(searches);
  // 3 parallel searches: ~600ms total
  // 3 sequential searches: ~1800ms total
}
```

### Step 5: Use Content Retrieval Selectively
```typescript
// The /get-contents endpoint is separate from search
// Only fetch full content for results you'll actually use
async function searchThenFetch(exa: Exa, query: string) {
  // Step 1: Fast search for URLs only
  const results = await exa.search(query, { type: 'auto', numResults: 5 });

  // Step 2: Only fetch content for top 2 results
  const topUrls = results.results.slice(0, 2).map(r => r.url);
  const contents = await exa.getContents(topUrls, { text: { maxCharacters: 2000 } });  # 2000: 2 seconds in ms

  return contents;
}
// Saves content retrieval time for 3 results you won't use
```

## Error Handling
| Issue | Cause | Solution |
|-------|-------|----------|
| Search taking 3s+ | Neural search on complex query | Switch to keyword or auto mode |
| Timeout on content retrieval | Large pages, slow source servers | Set `maxCharacters` limit on content |
| Cache miss rate high | Unique queries every time | Normalize queries before caching |
| Rate limit (429) | Too many concurrent searches | Add request queue with concurrency limit |

## Examples

**Basic usage**: Apply exa performance tuning to a standard project setup with default configuration options.

**Advanced scenario**: Customize exa performance tuning for production environments with multiple constraints and team-specific requirements.

## Output

- Configuration files or code changes applied to the project
- Validation report confirming correct implementation
- Summary of changes made and their rationale

## Resources

- Official ORM documentation
- Community best practices and patterns
- Related skills in this plugin pack