Error Handling in Tools
Build resilient AI agents through robust error handling and graceful degradation
Your Progress
0 / 5 completedRecovery Strategies: Bouncing Back
Once you've categorized an error, you need a recovery strategy. The right strategy depends on the error type—retry transient failures, use fallbacks for service outages, fail fast for permanent errors.
Interactive: Exponential Backoff Simulator
Watch how retry attempts increase wait time between tries
Recovery Strategy Comparison
🔄
Retry with Backoff
Exponentially increase wait time between retries
When to use:
Transient errors (network, timeout)
Example:
1s → 2s → 4s → 8s
✓ Handles temporary issues automatically
⚠ Can delay results, may waste resources
🔀
Fallback/Circuit Breaker
Switch to alternative tool or cached data
When to use:
Service degradation, repeated failures
Example:
Primary API down → Use secondary API
✓ Maintains functionality during outages
⚠ Requires alternative data sources
⬇️
Graceful Degradation
Return partial results, skip failed components
When to use:
Non-critical feature failures
Example:
Image generation fails → Return text-only
✓ User gets some value vs total failure
⚠ Reduced functionality may confuse users
⚡
Fail Fast
Return error immediately without retry
When to use:
Permanent errors, validation failures
Example:
404 Not Found → Return error message
✓ Quick feedback, no wasted resources
⚠ No recovery for transient issues
Retry Best Practices
Set Maximum Retries
Prevent infinite retry loops
Implementation:
Max 3-5 attempts before giving up
Use Exponential Backoff
Give systems time to recover
Implementation:
1s, 2s, 4s, 8s between attempts
Add Jitter
Prevent thundering herd
Implementation:
delay = baseDelay * 2^n + random(0, 1000ms)
Log All Attempts
Essential for debugging patterns
Implementation:
Log: attempt, delay, error, outcome