fix: surface actual API error messages instead of generic NoOutputGeneratedError

src/api/providers/anthropic-vertex.ts

@@ -175,6 +175,7 @@ export class AnthropicVertexHandler extends BaseProvider implements SingleComple
 		try {
 			const result = streamText(requestOptions)
 
+			let lastStreamError: string | undefined
 			for await (const part of result.fullStream) {
 				// Capture thinking signature from stream events
 				// The AI SDK's @ai-sdk/anthropic emits the signature as a reasoning-delta
@@ -193,15 +194,25 @@ export class AnthropicVertexHandler extends BaseProvider implements SingleComple
 				}
 
 				for (const chunk of processAiSdkStreamPart(part)) {
+					if (chunk.type === "error") {
+						lastStreamError = chunk.message
+					}
 					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, modelConfig.info, providerMetadata)
+			try {
+				const usage = await result.usage
+				const providerMetadata = await result.providerMetadata
+				if (usage) {
+					yield this.processUsageMetrics(usage, modelConfig.info, providerMetadata)
+				}
+			} catch (usageError) {
+				if (lastStreamError) {
+					throw new Error(lastStreamError)
+				}
+				throw usageError
 			}
 		} catch (error) {
 			const errorMessage = error instanceof Error ? error.message : String(error)

src/api/providers/anthropic.ts

@@ -151,6 +151,7 @@ export class AnthropicHandler extends BaseProvider implements SingleCompletionHa
 		try {
 			const result = streamText(requestOptions)
 
+			let lastStreamError: string | undefined
 			for await (const part of result.fullStream) {
 				// Capture thinking signature from stream events
 				// The AI SDK's @ai-sdk/anthropic emits the signature as a reasoning-delta
@@ -169,15 +170,25 @@ export class AnthropicHandler extends BaseProvider implements SingleCompletionHa
 				}
 
 				for (const chunk of processAiSdkStreamPart(part)) {
+					if (chunk.type === "error") {
+						lastStreamError = chunk.message
+					}
 					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, modelConfig.info, providerMetadata)
+			try {
+				const usage = await result.usage
+				const providerMetadata = await result.providerMetadata
+				if (usage) {
+					yield this.processUsageMetrics(usage, modelConfig.info, providerMetadata)
+				}
+			} catch (usageError) {
+				if (lastStreamError) {
+					throw new Error(lastStreamError)
+				}
+				throw usageError
 			}
 		} catch (error) {
 			const errorMessage = error instanceof Error ? error.message : String(error)

src/api/providers/azure.ts

@@ -9,7 +9,7 @@ import type { ApiHandlerOptions } from "../../shared/api"
 import {
 	convertToAiSdkMessages,
 	convertToolsForAiSdk,
-	processAiSdkStreamPart,
+	consumeAiSdkStream,
 	mapToolChoice,
 	handleAiSdkError,
 } from "../transform/ai-sdk"
@@ -159,19 +159,11 @@ export class AzureHandler extends BaseProvider implements SingleCompletionHandle
 		const result = streamText(requestOptions)
 
 		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)
-			}
+			const processUsage = this.processUsageMetrics.bind(this)
+			yield* consumeAiSdkStream(result, async function* () {
+				const [usage, providerMetadata] = await Promise.all([result.usage, result.providerMetadata])
+				yield processUsage(usage, providerMetadata as Parameters<typeof processUsage>[1])
+			})
 		} catch (error) {
 			// Handle AI SDK errors (AI_RetryError, AI_APICallError, etc.)
 			throw handleAiSdkError(error, "Azure AI Foundry")

src/api/providers/baseten.ts

@@ -9,7 +9,7 @@ import type { ApiHandlerOptions } from "../../shared/api"
 import {
 	convertToAiSdkMessages,
 	convertToolsForAiSdk,
-	processAiSdkStreamPart,
+	consumeAiSdkStream,
 	mapToolChoice,
 	handleAiSdkError,
 } from "../transform/ai-sdk"
@@ -118,16 +118,11 @@ export class BasetenHandler extends BaseProvider implements SingleCompletionHand
 		const result = streamText(requestOptions)
 
 		try {
-			for await (const part of result.fullStream) {
-				for (const chunk of processAiSdkStreamPart(part)) {
-					yield chunk
-				}
-			}
-
-			const usage = await result.usage
-			if (usage) {
-				yield this.processUsageMetrics(usage)
-			}
+			const processUsage = this.processUsageMetrics.bind(this)
+			yield* consumeAiSdkStream(result, async function* () {
+				const usage = await result.usage
+				yield processUsage(usage)
+			})
 		} catch (error) {
 			throw handleAiSdkError(error, "Baseten")
 		}

src/api/providers/bedrock.ts

@@ -343,6 +343,8 @@ export class AwsBedrockHandler extends BaseProvider implements SingleCompletionH
 		try {
 			const result = streamText(requestOptions)
 
+			let lastStreamError: string | undefined
+
 			// Process the full stream
 			for await (const part of result.fullStream) {
 				// Capture thinking signature from stream events.
@@ -371,15 +373,25 @@ export class AwsBedrockHandler extends BaseProvider implements SingleCompletionH
 				}
 
 				for (const chunk of processAiSdkStreamPart(part)) {
+					if (chunk.type === "error") {
+						lastStreamError = chunk.message
+					}
 					yield chunk
 				}
 			}
 
 			// Yield usage metrics at the end
-			const usage = await result.usage
-			const providerMetadata = await result.providerMetadata
-			if (usage) {
-				yield this.processUsageMetrics(usage, modelConfig.info, providerMetadata)
+			try {
+				const usage = await result.usage
+				const providerMetadata = await result.providerMetadata
+				if (usage) {
+					yield this.processUsageMetrics(usage, modelConfig.info, providerMetadata)
+				}
+			} catch (usageError) {
+				if (lastStreamError) {
+					throw new Error(lastStreamError)
+				}
+				throw usageError
 			}
 		} catch (error) {
 			const errorMessage = error instanceof Error ? error.message : String(error)

src/api/providers/deepseek.ts

@@ -9,7 +9,7 @@ import type { ApiHandlerOptions } from "../../shared/api"
 import {
 	convertToAiSdkMessages,
 	convertToolsForAiSdk,
-	processAiSdkStreamPart,
+	consumeAiSdkStream,
 	mapToolChoice,
 	handleAiSdkError,
 } from "../transform/ai-sdk"
@@ -137,21 +137,12 @@ export class DeepSeekHandler extends BaseProvider implements SingleCompletionHan
 		const result = streamText(requestOptions)
 
 		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)
-			}
+			const processUsage = this.processUsageMetrics.bind(this)
+			yield* consumeAiSdkStream(result, async function* () {
+				const [usage, providerMetadata] = await Promise.all([result.usage, result.providerMetadata])
+				yield processUsage(usage, providerMetadata as Parameters<typeof processUsage>[1])
+			})
 		} catch (error) {
-			// Handle AI SDK errors (AI_RetryError, AI_APICallError, etc.)
 			throw handleAiSdkError(error, "DeepSeek")
 		}
 	}

src/api/providers/fireworks.ts

@@ -9,7 +9,7 @@ import type { ApiHandlerOptions } from "../../shared/api"
 import {
 	convertToAiSdkMessages,
 	convertToolsForAiSdk,
-	processAiSdkStreamPart,
+	consumeAiSdkStream,
 	mapToolChoice,
 	handleAiSdkError,
 } from "../transform/ai-sdk"
@@ -137,21 +137,12 @@ export class FireworksHandler extends BaseProvider implements SingleCompletionHa
 		const result = streamText(requestOptions)
 
 		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)
-			}
+			const processUsage = this.processUsageMetrics.bind(this)
+			yield* consumeAiSdkStream(result, async function* () {
+				const [usage, providerMetadata] = await Promise.all([result.usage, result.providerMetadata])
+				yield processUsage(usage, providerMetadata as Parameters<typeof processUsage>[1])
+			})
 		} catch (error) {
-			// Handle AI SDK errors (AI_RetryError, AI_APICallError, etc.)
 			throw handleAiSdkError(error, "Fireworks")
 		}
 	}

src/api/providers/gemini.ts

@@ -133,6 +133,7 @@ export class GeminiHandler extends BaseProvider implements SingleCompletionHandl
 
 			// Track whether any text content was yielded (not just reasoning/thinking)
 			let hasContent = false
+			let lastStreamError: string | undefined
 
 			// Process the full stream to get all events including reasoning
 			for await (const part of result.fullStream) {
@@ -146,6 +147,9 @@ export class GeminiHandler extends BaseProvider implements SingleCompletionHandl
 				}
 
 				for (const chunk of processAiSdkStreamPart(part)) {
+					if (chunk.type === "error") {
+						lastStreamError = chunk.message
+					}
 					if (chunk.type === "text" || chunk.type === "tool_call_start") {
 						hasContent = true
 					}
@@ -163,7 +167,15 @@ export class GeminiHandler extends BaseProvider implements SingleCompletionHandl
 			}
 
 			// Extract grounding sources from providerMetadata if available
-			const providerMetadata = await result.providerMetadata
+			let providerMetadata: Awaited<typeof result.providerMetadata>
+			try {
+				providerMetadata = await result.providerMetadata
+			} catch (metaError) {
+				if (lastStreamError) {
+					throw new Error(lastStreamError)
+				}
+				throw metaError
+			}
 			const groundingMetadata = providerMetadata?.google as
 				| {
 						groundingMetadata?: {
@@ -190,6 +202,9 @@ export class GeminiHandler extends BaseProvider implements SingleCompletionHandl
 					yield this.processUsageMetrics(usage, info, providerMetadata)
 				}
 			} catch (usageError) {
+				if (lastStreamError) {
+					throw new Error(lastStreamError)
+				}
 				if (usageError instanceof NoOutputGeneratedError) {
 					// If we already yielded the empty-stream message, suppress this error
 					if (hasContent) {

src/api/providers/lm-studio.ts

@@ -8,7 +8,7 @@ import type { ApiHandlerOptions } from "../../shared/api"
 import {
 	convertToAiSdkMessages,
 	convertToolsForAiSdk,
-	processAiSdkStreamPart,
+	consumeAiSdkStream,
 	mapToolChoice,
 	handleAiSdkError,
 } from "../transform/ai-sdk"
@@ -79,16 +79,11 @@ export class LmStudioHandler extends OpenAICompatibleHandler implements SingleCo
 		const result = streamText(requestOptions)
 
 		try {
-			for await (const part of result.fullStream) {
-				for (const chunk of processAiSdkStreamPart(part)) {
-					yield chunk
-				}
-			}
-
-			const usage = await result.usage
-			if (usage) {
-				yield this.processUsageMetrics(usage)
-			}
+			const processUsage = this.processUsageMetrics.bind(this)
+			yield* consumeAiSdkStream(result, async function* () {
+				const usage = await result.usage
+				yield processUsage(usage)
+			})
 		} catch (error) {
 			throw handleAiSdkError(error, "LM Studio")
 		}

src/api/providers/minimax.ts

@@ -128,6 +128,8 @@ export class MiniMaxHandler extends BaseProvider implements SingleCompletionHand
 
 		try {
 			const result = streamText(requestOptions as Parameters<typeof streamText>[0])
+
+			let lastStreamError: string | undefined
 
 			for await (const part of result.fullStream) {
 				const anthropicMetadata = (
@@ -153,14 +155,24 @@ export class MiniMaxHandler extends BaseProvider implements SingleCompletionHand
 				}
 
 				for (const chunk of processAiSdkStreamPart(part)) {
+					if (chunk.type === "error") {
+						lastStreamError = chunk.message
+					}
 					yield chunk
 				}
 			}
 
-			const usage = await result.usage
-			const providerMetadata = await result.providerMetadata
-			if (usage) {
-				yield this.processUsageMetrics(usage, modelConfig.info, providerMetadata)
+			try {
+				const usage = await result.usage
+				const providerMetadata = await result.providerMetadata
+				if (usage) {
+					yield this.processUsageMetrics(usage, modelConfig.info, providerMetadata)
+				}
+			} catch (usageError) {
+				if (lastStreamError) {
+					throw new Error(lastStreamError)
+				}
+				throw usageError
 			}
 		} catch (error) {
 			throw handleAiSdkError(error, this.providerName)