Exa Incident Runbook

Exa Incident Runbook

Prerequisites

• An incident ID, commander, affected environment/service, approved communication route, and redacted diagnostic access.
• A safe mitigation/rollback path plus data/policy owner for query/result exposure decisions.

Instructions

1. Declare scope and severity, preserve minimal redacted evidence, and stabilize customer or automation impact.
2. Use the documented pause, credential revocation, rate reduction, or fallback before deep diagnosis.
3. Verify recovery against the SLO and policy boundary, then create root-cause/prevention follow-up work.

Output

• A time-stamped incident record with scope, mitigation, evidence, owner, and verified recovery or escalation.

Examples

For an unexpected result or availability incident, record opaque request IDs, environment, status/latency aggregates, and policy category, then reduce traffic or use the approved fallback. Do not share raw customer queries, result content, or credentials in incident channels.

Overview

Rapid incident response procedures for Exa search API issues. Exa errors include a requestId field for support escalation. Default rate limit is 10 QPS. Contact hello@exa.ai for urgent production issues.

Severity Levels

Quick Triage (Run First)

set -euo pipefail
echo "=== Exa Triage ==="

# 1. Test API connectivity
echo -n "API Status: "
HTTP_CODE=$(curl -s -o /tmp/exa-triage.json -w "%{http_code}" \
  -X POST https://api.exa.ai/search \
  -H "x-api-key: $EXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query":"triage test","numResults":1}')
echo "$HTTP_CODE"

# 2. Show error details if not 200
if [ "$HTTP_CODE" != "200" ]; then
  echo "Error response:"
  cat /tmp/exa-triage.json | python3 -m json.tool 2>/dev/null || cat /tmp/exa-triage.json
fi

# 3. Check if it's a key issue
echo ""
echo "API Key: ${EXA_API_KEY:+SET (${#EXA_API_KEY} chars)}"

Decision Tree

Exa API returning errors?
├── YES: What HTTP code?
│   ├── 401 → API key invalid/expired → Regenerate at dashboard.exa.ai
│   ├── 402 → Credits exhausted → Top up at dashboard.exa.ai
│   ├── 429 → Rate limited → Implement backoff, enable caching
│   ├── 5xx → Exa server issue → Retry with backoff, wait for resolution
│   └── 400 → Bad request → Fix request parameters
└── NO: Is search quality degraded?
    ├── Empty results → Broaden query, check date/domain filters
    ├── Low relevance → Switch search type, rephrase query
    └── Slow responses → Switch to faster search type, add caching

Immediate Actions by Error Code

401/403 — Authentication

set -euo pipefail
# Verify API key
echo "Key present: ${EXA_API_KEY:+yes}"
echo "Key length: ${#EXA_API_KEY}"

# Test with a simple search
curl -v -X POST https://api.exa.ai/search \
  -H "x-api-key: $EXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query":"auth test","numResults":1}' 2>&1 | grep "< HTTP"

# Fix: regenerate key at dashboard.exa.ai and update env

429 — Rate Limited

// Enable emergency caching to reduce API calls
import { LRUCache } from "lru-cache";

const emergencyCache = new LRUCache<string, any>({
  max: 10000,
  ttl: 30 * 60 * 1000, // 30-minute emergency TTL
});

// Reduce concurrent requests
import PQueue from "p-queue";
const queue = new PQueue({ concurrency: 3, interval: 1000, intervalCap: 5 });

5xx — Exa Server Errors

// Enable graceful degradation
async function searchWithFallback(query: string, opts: any) {
  try {
    return await exa.searchAndContents(query, opts);
  } catch (err: any) {
    if (err.status >= 500) {
      console.error(`[Exa] ${err.status}: ${err.message} (requestId: ${err.requestId})`);
      // Return cached results or show degraded UI
      const cached = emergencyCache.get(query);
      if (cached) return cached;
      return { results: [], _degraded: true };
    }
    throw err;
  }
}

Communication Templates

Internal (Slack)

P[1-4] INCIDENT: Exa Search Integration
Status: INVESTIGATING
Impact: [Describe user impact]
Error: [HTTP code] [error tag]
RequestId: [from error response]
Current action: [What you're doing]
Next update: [Time]

Support Escalation

To: hello@exa.ai
Subject: [P1/P2] Production issue — [brief description]

RequestId: [from error response]
Timestamp: [ISO 8601]
HTTP Status: [code]
Error Tag: [tag from response]
Frequency: [every request / intermittent / percentage]
Impact: [number of affected users/requests]

Post-Incident

Evidence Collection

set -euo pipefail
# Capture recent error logs
kubectl logs -l app=exa-integration --since=1h 2>/dev/null | grep -i "error\|429\|500" | tail -50

# Capture metrics snapshot
curl -s "localhost:9090/api/v1/query?query=rate(exa_search_error[1h])" 2>/dev/null

Postmortem Template

## Incident: Exa [Error Type]
**Date:** YYYY-MM-DD | **Duration:** Xh Ym | **Severity:** P[1-4]

### Summary
[1-2 sentence description]

### Timeline
- HH:MM — First error detected
- HH:MM — Triage began
- HH:MM — Root cause identified
- HH:MM — Mitigation applied
- HH:MM — Full recovery

### Root Cause
[Technical explanation]

### Action Items
- [ ] [Preventive measure] — Owner — Due date

Error Handling

Resources

• Exa Error Codes
• Exa Support

Next Steps

For data handling, see exa-data-handling. For debugging, see exa-debug-bundle.