AI Provider Chaos

Simulate realistic AI provider failures including rate limits, overloaded models, and API errors.

Provider-Specific Errors

Test different provider error conditions:

Rate Limit
Model Overloaded
Context Length
Content Filter

Simulate rate limiting:

import { openai } from '@ai-sdk/openai'
import { generateText } from 'ai'
import { cruelModel } from 'cruel/ai-sdk'

const model = cruelModel(openai('gpt-4o'), {
  rateLimit: 0.2 // 20% rate limit
})

try {
  const result = await generateText({
    model,
    prompt: 'Hello'
  })
} catch (error) {
  // Error: API call failed with status 429
  console.error('Rate limited:', error.message)
}

Simulate overloaded models:

const model = cruelModel(openai('gpt-4o'), {
  overloaded: 0.1 // 10% overload
})

try {
  await generateText({ model, prompt: 'Test' })
} catch (error) {
  // Error: API call failed with status 529
  console.error('Model overloaded')
}

Simulate context length exceeded:

const model = cruelModel(openai('gpt-4o'), {
  contextLength: 0.05 // 5% context error
})

try {
  await generateText({
    model,
    prompt: 'Very long prompt...'.repeat(1000)
  })
} catch (error) {
  // Error: Context length exceeded
  console.error('Prompt too long')
}

Simulate content filtering:

const model = cruelModel(openai('gpt-4o'), {
  contentFilter: 0.02 // 2% filter
})

try {
  await generateText({
    model,
    prompt: 'Potentially filtered content'
  })
} catch (error) {
  // Error: Content filter triggered
  console.error('Content blocked')
}

Model Unavailability

Simulate model downtime:

import { anthropic } from '@ai-sdk/anthropic'
import { generateText } from 'ai'
import { cruelModel } from 'cruel/ai-sdk'

const model = cruelModel(anthropic('claude-3-5-sonnet-20241022'), {
  modelUnavailable: 0.15 // 15% unavailable
})

try {
  const result = await generateText({
    model,
    prompt: 'Explain distributed systems'
  })
  console.log(result.text)
} catch (error) {
  // Error: Model not available
  console.error('Model unavailable, try fallback')
}

Authentication Errors

const model = cruelModel(openai('gpt-4o'), {
  invalidApiKey: 0.01 // 1% invalid key
})

try {
  await generateText({ model, prompt: 'Test' })
} catch (error) {
  // Error: Invalid API key
}

Gateway Integration Example

Test with AI Gateway under chaos:

import { gateway } from '@ai-sdk/gateway'
import { generateText } from 'ai'
import { cruelModel } from 'cruel/ai-sdk'

const model = cruelModel(gateway('google/gemini-2.0-flash-exp'), {
  rateLimit: 0.1,
  overloaded: 0.1,
  delay: [200, 1500],
  partialResponse: 0.2,
  slowTokens: [50, 300],
  onChaos: (event) => {
    console.log(`[chaos] ${event.type}`, event.modelId)
  }
})

let successes = 0
let failures = 0

for (let i = 0; i < 10; i++) {
  const start = Date.now()
  try {
    const result = await generateText({
      model,
      prompt: `Request ${i + 1}: Name a color.`,
      maxRetries: 2
    })
    successes++
    console.log(`[${i + 1}] ${Date.now() - start}ms:`, result.text.slice(0, 50))
  } catch (e) {
    failures++
    console.log(`[${i + 1}] ${Date.now() - start}ms failed:`, (e as Error).message)
  }
}

console.log(`\nsuccesses: ${successes}, failures: ${failures}`)

Multi-Provider Testing

Test fallback between providers:

import { openai } from '@ai-sdk/openai'
import { anthropic } from '@ai-sdk/anthropic'
import { google } from '@ai-sdk/google'
import { generateText } from 'ai'
import { cruelModel, presets } from 'cruel/ai-sdk'

const providers = [
  cruelModel(openai('gpt-4o'), presets.unstable),
  cruelModel(anthropic('claude-3-5-sonnet-20241022'), presets.unstable),
  cruelModel(google('gemini-2.0-flash-exp'), presets.unstable)
]

async function generateWithFallback(prompt: string) {
  for (let i = 0; i < providers.length; i++) {
    try {
      const result = await generateText({
        model: providers[i],
        prompt
      })
      console.log(`Success with provider ${i + 1}`)
      return result.text
    } catch (error) {
      console.log(`Provider ${i + 1} failed:`, error.message)
      if (i === providers.length - 1) throw error
    }
  }
}

const text = await generateWithFallback('Write a haiku about reliability')
console.log(text)

Partial Responses

Handle incomplete responses:

import { openai } from '@ai-sdk/openai'
import { generateText } from 'ai'
import { cruelModel } from 'cruel/ai-sdk'

const model = cruelModel(openai('gpt-4o'), {
  partialResponse: 0.3 // 30% partial
})

const result = await generateText({
  model,
  prompt: 'Write a long essay about testing (at least 500 words)'
})

if (result.text.length < 500) {
  console.log('Got partial response:', result.text.length, 'chars')
  console.log('Finish reason:', result.finishReason)
  // Handle incomplete response
}

Custom Finish Reasons

Override finish reasons for testing:

import { cruelModel } from 'cruel/ai-sdk'
import { openai } from '@ai-sdk/openai'
import { generateText } from 'ai'

const model = cruelModel(openai('gpt-4o'), {
  finishReason: 'length' // Force length limit
})

const result = await generateText({
  model,
  prompt: 'Count to 100'
})

console.log('Finish reason:', result.finishReason)
// Always 'length' instead of 'stop'

if (result.finishReason === 'length') {
  console.log('Response was truncated')
}

Token Usage Chaos

Override token usage for testing:

const model = cruelModel(openai('gpt-4o'), {
  tokenUsage: {
    inputTokens: 10000,  // Simulate large input
    outputTokens: 5000   // Simulate large output
  }
})

const result = await generateText({
  model,
  prompt: 'Hello'
})

console.log('Usage:', result.usage)
// { promptTokens: 10000, completionTokens: 5000, totalTokens: 15000 }

if (result.usage.totalTokens > 8000) {
  console.log('Warning: High token usage')
}

Testing with Jest

Comprehensive provider resilience tests:

import { describe, test, expect, jest } from '@jest/globals'
import { openai } from '@ai-sdk/openai'
import { generateText } from 'ai'
import { cruelModel } from 'cruel/ai-sdk'

describe('AI Provider Resilience', () => {
  test('retries on rate limit', async () => {
    let attempts = 0
    
    const model = cruelModel(openai('gpt-4o'), {
      rateLimit: 0.5
    })

    for (let i = 0; i < 3; i++) {
      try {
        attempts++
        await generateText({ model, prompt: 'Test' })
        break
      } catch (error) {
        if (i === 2) throw error
        await new Promise(r => setTimeout(r, 1000))
      }
    }

    expect(attempts).toBeGreaterThanOrEqual(1)
  })

  test('handles overloaded model', async () => {
    const model = cruelModel(openai('gpt-4o'), {
      overloaded: 1
    })

    await expect(generateText({
      model,
      prompt: 'Test'
    })).rejects.toThrow(/529|overload/i)
  })

  test('validates context length', async () => {
    const model = cruelModel(openai('gpt-4o'), {
      contextLength: 1
    })

    await expect(generateText({
      model,
      prompt: 'Test'
    })).rejects.toThrow(/context|length/i)
  })

  test('handles content filter', async () => {
    const model = cruelModel(openai('gpt-4o'), {
      contentFilter: 1
    })

    await expect(generateText({
      model,
      prompt: 'Test'
    })).rejects.toThrow(/content|filter/i)
  })

  test('falls back on model unavailable', async () => {
    const primary = cruelModel(openai('gpt-4o'), {
      modelUnavailable: 1
    })

    const fallback = cruelModel(openai('gpt-4o-mini'), {
      modelUnavailable: 0
    })

    let result
    try {
      result = await generateText({ model: primary, prompt: 'Test' })
    } catch {
      result = await generateText({ model: fallback, prompt: 'Test' })
    }

    expect(result).toBeDefined()
    expect(result.text).toBeTruthy()
  })
})

Production Monitoring

Monitor provider health in production:

import { cruelModel, diagnostics } from 'cruel/ai-sdk'
import { openai } from '@ai-sdk/openai'
import { generateText } from 'ai'

class ProviderMonitor {
  private ctx = diagnostics.context()
  private requestId = 0

  async generate(prompt: string) {
    const id = ++this.requestId
    const track = diagnostics.tracker(this.ctx)

    const model = cruelModel(openai('gpt-4o'), {
      // Only in development
      ...( process.env.NODE_ENV === 'development' && {
        rateLimit: 0.1,
        overloaded: 0.05
      }),
      onChaos: track
    })

    diagnostics.before(this.ctx, id)
    const start = Date.now()

    try {
      const result = await generateText({ model, prompt })
      diagnostics.success(this.ctx, id, Date.now() - start, result.text)
      return result
    } catch (error) {
      diagnostics.failure(this.ctx, id, Date.now() - start, error)
      throw error
    }
  }

  getHealthMetrics() {
    const stats = diagnostics.stats(this.ctx)
    return {
      totalRequests: stats.total,
      successRate: stats.successRate,
      avgLatency: stats.latency.success.avg,
      p95Latency: stats.latency.success.p95,
      commonErrors: stats.errors.slice(0, 3),
      chaosEvents: stats.events
    }
  }
}

const monitor = new ProviderMonitor()

for (let i = 0; i < 20; i++) {
  try {
    await monitor.generate('Test prompt')
  } catch {}
}

console.log('Health:', monitor.getHealthMetrics())

Basic Usage

AI Applications

Patterns

AI Provider Chaos

AI Provider Chaos

Provider-Specific Errors

Model Unavailability

Authentication Errors

Gateway Integration Example

Multi-Provider Testing

Partial Responses

Custom Finish Reasons

Token Usage Chaos

Testing with Jest

Production Monitoring

Next Steps

Build docs developers (and LLMs) love

Basic Usage

AI Applications

Patterns

Documentation Index

​AI Provider Chaos

​Provider-Specific Errors

​Model Unavailability

​Authentication Errors

​Gateway Integration Example

​Multi-Provider Testing

​Partial Responses

​Custom Finish Reasons

​Token Usage Chaos

​Testing with Jest

​Production Monitoring

​Next Steps

Build docs developers (and LLMs) love

AI Provider Chaos

Provider-Specific Errors

Model Unavailability

Authentication Errors

Gateway Integration Example

Multi-Provider Testing

Partial Responses

Custom Finish Reasons

Token Usage Chaos

Testing with Jest

Production Monitoring

Next Steps