Error Recovery

Robust error handling is critical for production AI applications. The AI SDK provides multiple patterns for error detection, handling, and recovery.

Error Types

AI SDK Error Classes

The AI SDK provides specific error classes:

import {
  APICallError,
  InvalidArgumentError,
  InvalidToolInputError,
  NoSuchToolError,
  NoObjectGeneratedError,
  TypeValidationError,
} from 'ai';

try {
  const result = await generateText({
    model: openai('gpt-4'),
    prompt: 'Hello',
  });
} catch (error) {
  if (APICallError.isInstance(error)) {
    console.error('API call failed:', error.message);
    console.error('Status code:', error.statusCode);
  } else if (InvalidArgumentError.isInstance(error)) {
    console.error('Invalid argument:', error.message);
  }
}

Common Error Scenarios

import {
  APICallError,
  InvalidToolInputError,
  NoObjectGeneratedError,
} from 'ai';

// API rate limiting
if (APICallError.isInstance(error) && error.statusCode === 429) {
  console.log('Rate limited, retrying after delay...');
  await delay(5000);
}

// Invalid tool parameters
if (InvalidToolInputError.isInstance(error)) {
  console.error('Tool received invalid input:', error.toolInput);
  console.error('For tool:', error.toolName);
}

// Object generation failed
if (NoObjectGeneratedError.isInstance(error)) {
  console.error('Failed to generate structured output');
  console.error('Response:', error.text);
}

Retry Strategies

Built-in Retry

Use the maxRetries option:

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

const result = await generateText({
  model: openai('gpt-4'),
  prompt: 'Hello',
  maxRetries: 3, // Retry up to 3 times on failure
});

Custom Retry Logic

Implement exponential backoff:

async function generateWithRetry<T>(
  fn: () => Promise<T>,
  options: {
    maxRetries?: number;
    baseDelay?: number;
    maxDelay?: number;
  } = {},
): Promise<T> {
  const {
    maxRetries = 3,
    baseDelay = 1000,
    maxDelay = 30000,
  } = options;

  let lastError: Error;

  for (let attempt = 0; attempt <= maxRetries; attempt++) {
    try {
      return await fn();
    } catch (error) {
      lastError = error as Error;

      if (attempt === maxRetries) {
        throw error;
      }

      // Calculate exponential backoff with jitter
      const delay = Math.min(
        baseDelay * Math.pow(2, attempt) + Math.random() * 1000,
        maxDelay,
      );

      console.log(`Attempt ${attempt + 1} failed, retrying in ${delay}ms...`);
      await new Promise(resolve => setTimeout(resolve, delay));
    }
  }

  throw lastError!;
}

// Usage
const result = await generateWithRetry(() =>
  generateText({
    model: openai('gpt-4'),
    prompt: 'Hello',
  }),
);

Selective Retry

Retry only on specific errors:

import { APICallError } from 'ai';

const RETRYABLE_STATUS_CODES = [408, 429, 500, 502, 503, 504];

async function generateWithSelectiveRetry<T>(
  fn: () => Promise<T>,
  maxRetries = 3,
): Promise<T> {
  for (let attempt = 0; attempt <= maxRetries; attempt++) {
    try {
      return await fn();
    } catch (error) {
      if (
        APICallError.isInstance(error) &&
        RETRYABLE_STATUS_CODES.includes(error.statusCode ?? 0)
      ) {
        if (attempt < maxRetries) {
          await new Promise(resolve => setTimeout(resolve, 1000 * (attempt + 1)));
          continue;
        }
      }
      throw error;
    }
  }
  throw new Error('Max retries exceeded');
}

Fallback Models

Fall back to alternative models:

import { generateText } from 'ai';
import { openai } from '@ai-sdk/openai';
import { anthropic } from '@ai-sdk/anthropic';

async function generateWithFallback(prompt: string) {
  const models = [
    { name: 'gpt-4', model: openai('gpt-4') },
    { name: 'gpt-3.5-turbo', model: openai('gpt-3.5-turbo') },
    { name: 'claude-3-5-sonnet', model: anthropic('claude-3-5-sonnet-20241022') },
  ];

  let lastError: Error | null = null;

  for (const { name, model } of models) {
    try {
      console.log(`Trying ${name}...`);
      return await generateText({ model, prompt });
    } catch (error) {
      console.error(`${name} failed:`, error);
      lastError = error as Error;
      continue;
    }
  }

  throw new Error(
    `All models failed. Last error: ${lastError?.message}`,
  );
}

Streaming Error Recovery

Handling Stream Errors

import { streamText } from 'ai';
import { openai } from '@ai-sdk/openai';

const result = streamText({
  model: openai('gpt-4'),
  prompt: 'Write a story',
});

try {
  for await (const chunk of result.textStream) {
    process.stdout.write(chunk);
  }
} catch (error) {
  console.error('Stream failed:', error);
  // Implement recovery logic
}

Partial Results on Error

import { streamText } from 'ai';
import { openai } from '@ai-sdk/openai';

const result = streamText({
  model: openai('gpt-4'),
  prompt: 'Write a story',
});

let partialText = '';

try {
  for await (const chunk of result.textStream) {
    partialText += chunk;
    process.stdout.write(chunk);
  }
} catch (error) {
  console.error('Stream interrupted');
  console.log('Partial result:', partialText);
  // Save or use partial result
}

Resumable Streams

async function resumableGenerate({
  prompt,
  onProgress,
}: {
  prompt: string;
  onProgress: (text: string) => void;
}) {
  let accumulated = '';
  let attempts = 0;
  const maxAttempts = 3;

  while (attempts < maxAttempts) {
    try {
      const result = streamText({
        model: openai('gpt-4'),
        prompt: accumulated
          ? `Continue from: ${accumulated}\n\nOriginal prompt: ${prompt}`
          : prompt,
      });

      for await (const chunk of result.textStream) {
        accumulated += chunk;
        onProgress(accumulated);
      }

      return accumulated;
    } catch (error) {
      attempts++;
      console.error(`Attempt ${attempts} failed:`, error);

      if (attempts >= maxAttempts) {
        throw new Error(`Failed after ${maxAttempts} attempts`);
      }

      await new Promise(resolve => setTimeout(resolve, 2000));
    }
  }
}

Tool Call Error Handling

Validating Tool Inputs

import { generateText, tool } from 'ai';
import { openai } from '@ai-sdk/openai';
import { z } from 'zod';

const result = await generateText({
  model: openai('gpt-4'),
  prompt: 'Get weather for London',
  tools: {
    getWeather: tool({
      description: 'Get weather for a location',
      inputSchema: z.object({
        location: z.string().min(1, 'Location is required'),
        units: z.enum(['celsius', 'fahrenheit']).default('celsius'),
      }),
      execute: async ({ location, units }) => {
        try {
          const weather = await fetchWeather(location, units);
          return weather;
        } catch (error) {
          return {
            error: true,
            message: `Failed to fetch weather: ${error.message}`,
          };
        }
      },
    }),
  },
});

Tool Execution Errors

import { generateText, tool } from 'ai';
import { openai } from '@ai-sdk/openai';
import { z } from 'zod';

const result = await generateText({
  model: openai('gpt-4'),
  tools: {
    riskyOperation: tool({
      inputSchema: z.object({ action: z.string() }),
      execute: async ({ action }) => {
        try {
          return await performAction(action);
        } catch (error) {
          // Return error as tool result
          return {
            success: false,
            error: error.message,
            suggestedAction: 'Try again with different parameters',
          };
        }
      },
    }),
  },
});

RSC Error Recovery

Streamable UI Errors

import { createStreamableUI } from '@ai-sdk/rsc';

export async function generateComponent() {
  const stream = createStreamableUI(<div>Loading...</div>);

  (async () => {
    try {
      const data = await fetchData();
      stream.update(<Success data={data} />);
      stream.done();
    } catch (error) {
      stream.error(error);
    }
  })();

  return stream.value;
}

Client-Side Error Boundaries

'use client';

import { ErrorBoundary } from 'react-error-boundary';

function ErrorFallback({ error, resetErrorBoundary }) {
  return (
    <div role="alert">
      <h2>Something went wrong</h2>
      <pre>{error.message}</pre>
      <button onClick={resetErrorBoundary}>Try again</button>
    </div>
  );
}

export function SafeAIComponent({ children }) {
  return (
    <ErrorBoundary
      FallbackComponent={ErrorFallback}
      onReset={() => {
        // Reset component state
      }}
    >
      {children}
    </ErrorBoundary>
  );
}

Graceful Degradation

import { streamUI } from '@ai-sdk/rsc';
import { openai } from '@ai-sdk/openai';

export async function generateResponse(prompt: string) {
  try {
    const result = await streamUI({
      model: openai('gpt-4'),
      prompt,
      text: ({ content }) => <div>{content}</div>,
    });
    return result.value;
  } catch (error) {
    console.error('AI generation failed:', error);
    // Return fallback UI
    return (
      <div>
        <p>Unable to generate AI response at this time.</p>
        <p>Please try again later.</p>
      </div>
    );
  }
}

Timeout Handling

AbortSignal

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

const controller = new AbortController();
const timeoutId = setTimeout(() => controller.abort(), 30000); // 30s timeout

try {
  const result = await generateText({
    model: openai('gpt-4'),
    prompt: 'Write a long story',
    abortSignal: controller.signal,
  });
  clearTimeout(timeoutId);
  return result;
} catch (error) {
  if (error.name === 'AbortError') {
    console.error('Request timed out');
  }
  throw error;
}

Timeout Wrapper

async function withTimeout<T>(
  fn: (signal: AbortSignal) => Promise<T>,
  timeout: number,
): Promise<T> {
  const controller = new AbortController();
  const timeoutId = setTimeout(() => controller.abort(), timeout);

  try {
    return await fn(controller.signal);
  } finally {
    clearTimeout(timeoutId);
  }
}

// Usage
const result = await withTimeout(
  signal =>
    generateText({
      model: openai('gpt-4'),
      prompt: 'Hello',
      abortSignal: signal,
    }),
  30000,
);

Logging and Monitoring

Error Tracking

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

async function generateWithTracking(prompt: string) {
  const startTime = Date.now();
  const requestId = crypto.randomUUID();

  try {
    const result = await generateText({
      model: openai('gpt-4'),
      prompt,
    });

    // Log success
    console.log({
      requestId,
      duration: Date.now() - startTime,
      tokens: result.usage,
      status: 'success',
    });

    return result;
  } catch (error) {
    // Log error
    console.error({
      requestId,
      duration: Date.now() - startTime,
      error: error.message,
      status: 'error',
    });

    // Send to error tracking service
    // await trackError({ requestId, error });

    throw error;
  }
}

Best Practices

Use specific error checks: Use ErrorClass.isInstance() instead of instanceof
Implement retry logic: Add exponential backoff for transient failures
Set timeouts: Prevent indefinite hangs with AbortSignal
Validate inputs: Catch errors early with schema validation
Provide fallbacks: Have alternative models or cached responses
Log errors: Track failures for debugging and monitoring
Handle partial results: Save progress in streaming scenarios
Use error boundaries: Isolate UI errors in React applications

Getting Started

Foundations

AI SDK Core

Agents

AI SDK UI

AI SDK RSC

Advanced

Troubleshooting

Error Recovery

Error Recovery

Error Types

AI SDK Error Classes

Common Error Scenarios

Retry Strategies

Built-in Retry

Custom Retry Logic

Selective Retry

Fallback Models

Streaming Error Recovery

Handling Stream Errors

Partial Results on Error

Resumable Streams

Tool Call Error Handling

Validating Tool Inputs

Tool Execution Errors

RSC Error Recovery

Streamable UI Errors

Client-Side Error Boundaries

Graceful Degradation

Timeout Handling

AbortSignal

Timeout Wrapper

Logging and Monitoring

Error Tracking

Best Practices

Next Steps

Getting Started

Foundations

AI SDK Core

Agents

AI SDK UI

AI SDK RSC

Advanced

Troubleshooting

​Error Recovery

​Error Types

​AI SDK Error Classes

​Common Error Scenarios

​Retry Strategies

​Built-in Retry

​Custom Retry Logic

​Selective Retry

​Fallback Models

​Streaming Error Recovery

​Handling Stream Errors

​Partial Results on Error

​Resumable Streams

​Tool Call Error Handling

​Validating Tool Inputs

​Tool Execution Errors

​RSC Error Recovery

​Streamable UI Errors

​Client-Side Error Boundaries

​Graceful Degradation

​Timeout Handling

​AbortSignal

​Timeout Wrapper

​Logging and Monitoring

​Error Tracking

​Best Practices

​Next Steps

Error Recovery

Error Types

AI SDK Error Classes

Common Error Scenarios

Retry Strategies

Built-in Retry

Custom Retry Logic

Selective Retry

Fallback Models

Streaming Error Recovery

Handling Stream Errors

Partial Results on Error

Resumable Streams

Tool Call Error Handling

Validating Tool Inputs

Tool Execution Errors

RSC Error Recovery

Streamable UI Errors

Client-Side Error Boundaries

Graceful Degradation

Timeout Handling

AbortSignal

Timeout Wrapper

Logging and Monitoring

Error Tracking

Best Practices

Next Steps