fix(api): add proper error handling for AI SDK providers

- Added extractAiSdkErrorMessage utility to extract user-friendly error
  messages from AI SDK errors (AI_RetryError, AI_APICallError)
- Added handleAiSdkError utility to wrap errors with provider name and
  preserve status codes for retry logic
- Updated all AI SDK providers (Fireworks, Groq, DeepSeek, Cerebras,
  OpenAI-compatible) with try/catch error handling
- Added comprehensive tests for error handling utilities

This ensures errors like 'AI_RetryError: Failed after 3 attempts. Last
error: Too Many Requests' are properly surfaced in the UI instead of
showing 'No output generated. Check the stream for errors.'

Author: daniel-lxs

src/api/providers/cerebras.ts

@@ -11,6 +11,7 @@ import {
 	convertToolsForAiSdk,
 	processAiSdkStreamPart,
 	mapToolChoice,
+	handleAiSdkError,
 } from "../transform/ai-sdk"
 import { ApiStream, ApiStreamUsageChunk } from "../transform/stream"
 import { getModelParams } from "../transform/model-params"
@@ -120,17 +121,22 @@ export class CerebrasHandler extends BaseProvider implements SingleCompletionHan
 		// Use streamText for streaming responses
 		const result = streamText(requestOptions)
 
-		// Process the full stream to get all events including reasoning
-		for await (const part of result.fullStream) {
-			for (const chunk of processAiSdkStreamPart(part)) {
-				yield chunk
+		try {
+			// Process the full stream to get all events including reasoning
+			for await (const part of result.fullStream) {
+				for (const chunk of processAiSdkStreamPart(part)) {
+					yield chunk
+				}
 			}
-		}
 
-		// Yield usage metrics at the end
-		const usage = await result.usage
-		if (usage) {
-			yield this.processUsageMetrics(usage)
+			// Yield usage metrics at the end
+			const usage = await result.usage
+			if (usage) {
+				yield this.processUsageMetrics(usage)
+			}
+		} catch (error) {
+			// Handle AI SDK errors (AI_RetryError, AI_APICallError, etc.)
+			throw handleAiSdkError(error, "Cerebras")
 		}
 	}
 

src/api/providers/deepseek.ts

@@ -11,6 +11,7 @@ import {
 	convertToolsForAiSdk,
 	processAiSdkStreamPart,
 	mapToolChoice,
+	handleAiSdkError,
 } from "../transform/ai-sdk"
 import { ApiStream, ApiStreamUsageChunk } from "../transform/stream"
 import { getModelParams } from "../transform/model-params"
@@ -129,18 +130,23 @@ export class DeepSeekHandler extends BaseProvider implements SingleCompletionHan
 		// Use streamText for streaming responses
 		const result = streamText(requestOptions)
 
-		// Process the full stream to get all events including reasoning
-		for await (const part of result.fullStream) {
-			for (const chunk of processAiSdkStreamPart(part)) {
-				yield chunk
+		try {
+			// Process the full stream to get all events including reasoning
+			for await (const part of result.fullStream) {
+				for (const chunk of processAiSdkStreamPart(part)) {
+					yield chunk
+				}
 			}
-		}
 
-		// Yield usage metrics at the end, including cache metrics from providerMetadata
-		const usage = await result.usage
-		const providerMetadata = await result.providerMetadata
-		if (usage) {
-			yield this.processUsageMetrics(usage, providerMetadata as any)
+			// Yield usage metrics at the end, including cache metrics from providerMetadata
+			const usage = await result.usage
+			const providerMetadata = await result.providerMetadata
+			if (usage) {
+				yield this.processUsageMetrics(usage, providerMetadata as any)
+			}
+		} catch (error) {
+			// Handle AI SDK errors (AI_RetryError, AI_APICallError, etc.)
+			throw handleAiSdkError(error, "DeepSeek")
 		}
 	}
 

src/api/providers/fireworks.ts

@@ -11,6 +11,7 @@ import {
 	convertToolsForAiSdk,
 	processAiSdkStreamPart,
 	mapToolChoice,
+	handleAiSdkError,
 } from "../transform/ai-sdk"
 import { ApiStream, ApiStreamUsageChunk } from "../transform/stream"
 import { getModelParams } from "../transform/model-params"
@@ -135,18 +136,23 @@ export class FireworksHandler extends BaseProvider implements SingleCompletionHa
 		// Use streamText for streaming responses
 		const result = streamText(requestOptions)
 
-		// Process the full stream to get all events including reasoning
-		for await (const part of result.fullStream) {
-			for (const chunk of processAiSdkStreamPart(part)) {
-				yield chunk
+		try {
+			// Process the full stream to get all events including reasoning
+			for await (const part of result.fullStream) {
+				for (const chunk of processAiSdkStreamPart(part)) {
+					yield chunk
+				}
 			}
-		}
 
-		// Yield usage metrics at the end, including cache metrics from providerMetadata
-		const usage = await result.usage
-		const providerMetadata = await result.providerMetadata
-		if (usage) {
-			yield this.processUsageMetrics(usage, providerMetadata as any)
+			// Yield usage metrics at the end, including cache metrics from providerMetadata
+			const usage = await result.usage
+			const providerMetadata = await result.providerMetadata
+			if (usage) {
+				yield this.processUsageMetrics(usage, providerMetadata as any)
+			}
+		} catch (error) {
+			// Handle AI SDK errors (AI_RetryError, AI_APICallError, etc.)
+			throw handleAiSdkError(error, "Fireworks")
 		}
 	}
 

src/api/providers/groq.ts

@@ -11,6 +11,7 @@ import {
 	convertToolsForAiSdk,
 	processAiSdkStreamPart,
 	mapToolChoice,
+	handleAiSdkError,
 } from "../transform/ai-sdk"
 import { ApiStream, ApiStreamUsageChunk } from "../transform/stream"
 import { getModelParams } from "../transform/model-params"
@@ -137,18 +138,23 @@ export class GroqHandler extends BaseProvider implements SingleCompletionHandler
 		// Use streamText for streaming responses
 		const result = streamText(requestOptions)
 
-		// Process the full stream to get all events including reasoning
-		for await (const part of result.fullStream) {
-			for (const chunk of processAiSdkStreamPart(part)) {
-				yield chunk
+		try {
+			// Process the full stream to get all events including reasoning
+			for await (const part of result.fullStream) {
+				for (const chunk of processAiSdkStreamPart(part)) {
+					yield chunk
+				}
 			}
-		}
 
-		// Yield usage metrics at the end, including cache metrics from providerMetadata
-		const usage = await result.usage
-		const providerMetadata = await result.providerMetadata
-		if (usage) {
-			yield this.processUsageMetrics(usage, providerMetadata as any)
+			// Yield usage metrics at the end, including cache metrics from providerMetadata
+			const usage = await result.usage
+			const providerMetadata = await result.providerMetadata
+			if (usage) {
+				yield this.processUsageMetrics(usage, providerMetadata as any)
+			}
+		} catch (error) {
+			// Handle AI SDK errors (AI_RetryError, AI_APICallError, etc.)
+			throw handleAiSdkError(error, "Groq")
 		}
 	}
 

src/api/providers/openai-compatible.ts

@@ -17,6 +17,7 @@ import {
 	convertToolsForAiSdk,
 	processAiSdkStreamPart,
 	mapToolChoice,
+	handleAiSdkError,
 } from "../transform/ai-sdk"
 import { ApiStream, ApiStreamUsageChunk } from "../transform/stream"
 
@@ -150,18 +151,23 @@ export abstract class OpenAICompatibleHandler extends BaseProvider implements Si
 		// Use streamText for streaming responses
 		const result = streamText(requestOptions)
 
-		// Process the full stream to get all events
-		for await (const part of result.fullStream) {
-			// Use the processAiSdkStreamPart utility to convert stream parts
-			for (const chunk of processAiSdkStreamPart(part)) {
-				yield chunk
+		try {
+			// Process the full stream to get all events
+			for await (const part of result.fullStream) {
+				// Use the processAiSdkStreamPart utility to convert stream parts
+				for (const chunk of processAiSdkStreamPart(part)) {
+					yield chunk
+				}
 			}
-		}
 
-		// Yield usage metrics at the end
-		const usage = await result.usage
-		if (usage) {
-			yield this.processUsageMetrics(usage)
+			// Yield usage metrics at the end
+			const usage = await result.usage
+			if (usage) {
+				yield this.processUsageMetrics(usage)
+			}
+		} catch (error) {
+			// Handle AI SDK errors (AI_RetryError, AI_APICallError, etc.)
+			throw handleAiSdkError(error, this.config.providerName)
 		}
 	}
 

src/api/transform/__tests__/ai-sdk.spec.ts

@@ -1,6 +1,13 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 import OpenAI from "openai"
-import { convertToAiSdkMessages, convertToolsForAiSdk, processAiSdkStreamPart, mapToolChoice } from "../ai-sdk"
+import {
+	convertToAiSdkMessages,
+	convertToolsForAiSdk,
+	processAiSdkStreamPart,
+	mapToolChoice,
+	extractAiSdkErrorMessage,
+	handleAiSdkError,
+} from "../ai-sdk"
 
 vitest.mock("ai", () => ({
 	tool: vitest.fn((t) => t),
@@ -531,4 +538,110 @@ describe("AI SDK conversion utilities", () => {
 			expect(result).toBeUndefined()
 		})
 	})
+
+	describe("extractAiSdkErrorMessage", () => {
+		it("should return 'Unknown error' for null/undefined", () => {
+			expect(extractAiSdkErrorMessage(null)).toBe("Unknown error")
+			expect(extractAiSdkErrorMessage(undefined)).toBe("Unknown error")
+		})
+
+		it("should extract message from AI_RetryError", () => {
+			const retryError = {
+				name: "AI_RetryError",
+				message: "Failed after 3 attempts",
+				errors: [new Error("Error 1"), new Error("Error 2"), new Error("Too Many Requests")],
+				lastError: { message: "Too Many Requests", status: 429 },
+			}
+
+			const result = extractAiSdkErrorMessage(retryError)
+			expect(result).toBe("Failed after 3 attempts (429): Too Many Requests")
+		})
+
+		it("should handle AI_RetryError without status", () => {
+			const retryError = {
+				name: "AI_RetryError",
+				message: "Failed after 2 attempts",
+				errors: [new Error("Error 1"), new Error("Connection failed")],
+				lastError: { message: "Connection failed" },
+			}
+
+			const result = extractAiSdkErrorMessage(retryError)
+			expect(result).toBe("Failed after 2 attempts: Connection failed")
+		})
+
+		it("should extract message from AI_APICallError", () => {
+			const apiError = {
+				name: "AI_APICallError",
+				message: "Rate limit exceeded",
+				status: 429,
+			}
+
+			const result = extractAiSdkErrorMessage(apiError)
+			expect(result).toBe("API Error (429): Rate limit exceeded")
+		})
+
+		it("should handle AI_APICallError without status", () => {
+			const apiError = {
+				name: "AI_APICallError",
+				message: "Connection timeout",
+			}
+
+			const result = extractAiSdkErrorMessage(apiError)
+			expect(result).toBe("Connection timeout")
+		})
+
+		it("should extract message from standard Error", () => {
+			const error = new Error("Something went wrong")
+			expect(extractAiSdkErrorMessage(error)).toBe("Something went wrong")
+		})
+
+		it("should convert non-Error to string", () => {
+			expect(extractAiSdkErrorMessage("string error")).toBe("string error")
+			expect(extractAiSdkErrorMessage({ custom: "object" })).toBe("[object Object]")
+		})
+	})
+
+	describe("handleAiSdkError", () => {
+		it("should wrap error with provider name", () => {
+			const error = new Error("API Error")
+			const result = handleAiSdkError(error, "Fireworks")
+
+			expect(result.message).toBe("Fireworks: API Error")
+		})
+
+		it("should preserve status code from AI_RetryError", () => {
+			const retryError = {
+				name: "AI_RetryError",
+				errors: [new Error("Too Many Requests")],
+				lastError: { message: "Too Many Requests", status: 429 },
+			}
+
+			const result = handleAiSdkError(retryError, "Groq")
+
+			expect(result.message).toContain("Groq:")
+			expect(result.message).toContain("429")
+			expect((result as any).status).toBe(429)
+		})
+
+		it("should preserve status code from AI_APICallError", () => {
+			const apiError = {
+				name: "AI_APICallError",
+				message: "Unauthorized",
+				status: 401,
+			}
+
+			const result = handleAiSdkError(apiError, "DeepSeek")
+
+			expect(result.message).toContain("DeepSeek:")
+			expect(result.message).toContain("401")
+			expect((result as any).status).toBe(401)
+		})
+
+		it("should preserve original error as cause", () => {
+			const originalError = new Error("Original error")
+			const result = handleAiSdkError(originalError, "Cerebras")
+
+			expect((result as any).cause).toBe(originalError)
+		})
+	})
 })