@tanstack/ai-client 0.26.0 → 0.28.0

package/src/chat-client.ts

@@ -9,6 +9,15 @@ import {
   restoreInboundChunk,
   tanstackMetadata,
 } from '@tanstack/ai/client'
+import {
+  ByokBlockedError,
+  ByokMissingError,
+  ByokUnresolvedProviderError,
+} from '@tanstack/ai/byok'
+import {
+  prepareResolvedByokHeaders,
+  resolveByokProviderId,
+} from './byok/resolve'
 import { createNoOpChatDevtoolsBridge } from './devtools-noop'
 import {
   fetcherToConnectionAdapter,
@@ -27,6 +36,7 @@ import type {
   RunAgentResumeItem,
   StreamChunk,
 } from '@tanstack/ai/client'
+import type { ByokClient } from './byok'
 import type {
   ChatHydrationResult,
   ConnectionAdapter,
@@ -97,6 +107,8 @@ type ChatClientUpdateOptionsWithoutContext<
   /** @deprecated Use `forwardedProps` instead. */
   body?: Record<string, any>
   forwardedProps?: Record<string, any>
+  byok?: ByokClient
+  byokProvider?: () => string | undefined
   tools?: TTools
   interrupts?: TInterrupts
   queue?: QueueOption
@@ -348,6 +360,8 @@ export class ChatClient<
   // merged on every send, with `forwardedProps` winning on key collision.
   private bodyOption: Record<string, any> = {}
   private forwardedPropsOption: Record<string, any> = {}
+  private byok: ByokClient | undefined
+  private byokProvider: (() => string | undefined) | undefined
   private context: TContext | undefined = undefined
   private pendingMessageBody: Record<string, any> | undefined = undefined
   private queueConfig: NormalizedQueueConfig
@@ -494,6 +508,8 @@ export class ChatClient<
     // winning on key collision.
     this.bodyOption = options.body || {}
     this.forwardedPropsOption = options.forwardedProps || {}
+    this.byok = options.byok
+    this.byokProvider = options.byokProvider
     this.context = options.context
     this.queueConfig = normalizeQueueOption(options.queue)
     this.connection = normalizeConnectionAdapter(resolveTransport(options))
@@ -1882,8 +1898,12 @@ export class ChatClient<
    *   - A MultimodalContent object with content array and optional custom ID
    * @param body - Optional body parameters to merge with the client's base body for this request.
    *               Uses shallow merge with per-message body taking priority.
-   * @param sendOptions - Per-call overrides, e.g. `{ whenBusy: 'interrupt' }` to
-   *                      override the configured queue policy for this one send.
+   * @param sendOptions - Per-call overrides. `{ whenBusy }` overrides the
+   *                      queue policy for this one send. `{ body }`
+   *                      shallow-merges with `body` and with the chat-level
+   *                      `body` / `forwardedProps`. `sendOptions.body` wins
+   *                      on key collisions. Framework hooks forward this
+   *                      object as their second argument.
    *
    * @example
    * ```ts
@@ -1893,9 +1913,13 @@ export class ChatClient<
    * // Text message with custom body params
    * await client.sendMessage('Hello!', { temperature: 0.7 })
    *
-   * // Per-call whenBusy override (body must still be the 2nd arg on ChatClient)
+   * // Per-call whenBusy override
    * await client.sendMessage('Urgent', undefined, { whenBusy: 'interrupt' })
    *
+   * // Per-call body via options. Same effect as the positional arg.
+   * // This is the shape the framework hooks (`useChat`, `injectChat`) forward.
+   * await client.sendMessage('Hello!', undefined, { body: { temperature: 0.7 } })
+   *
    * // Multimodal message with image
    * await client.sendMessage({
    *   content: [
@@ -1933,13 +1957,15 @@ export class ChatClient<
       )
     }
 
+    const resolvedBody = { ...body, ...sendOptions?.body }
+
     if (this.isSendBusy()) {
       const { action, id } = this.decideWhenBusy(content, sendOptions)
       if (action === 'drop') {
         return
       }
       if (action === 'queue') {
-        this.enqueueMessage(content, body, id)
+        this.enqueueMessage(content, resolvedBody, id)
         return
       }
       // 'interrupt': abort the current stream, then send now.
@@ -1956,7 +1982,7 @@ export class ChatClient<
     }
 
     try {
-      await this.deliverMessage(content, body)
+      await this.deliverMessage(content, resolvedBody)
     } finally {
       this.sendInFlight = false
     }
@@ -2191,7 +2217,7 @@ export class ChatClient<
       // order (later spreads win):
       //   1. Legacy `body` option (deprecated).
       //   2. Canonical `forwardedProps` option (wins over `body`).
-      //   3. Per-message `body` arg passed to `sendMessage` (highest).
+      //   3. Per-call body (`pendingMessageBody`: positional + sendOptions.body).
       // The AG-UI standard `threadId` is sent at the wire's top level for
       // run/conversation correlation, so we no longer auto-emit a separate
       // `conversationId` here — `chat({ threadId })` server-side covers the
@@ -2231,6 +2257,15 @@ export class ChatClient<
       // AG-UI servers consuming `RunAgentInput.tools[].parameters` expect
       // JSON Schema; sending a Standard Schema instance directly would
       // serialize to an unusable shape.
+      let byokHeaders: Record<string, string> | undefined
+      if (this.byok) {
+        const provider = resolveByokProviderId(
+          this.byokProvider,
+          mergedBody.provider,
+        )
+        byokHeaders = await prepareResolvedByokHeaders(this.byok, provider)
+      }
+
       const runContext = {
         threadId: resumeThreadId ?? this.threadId,
         runId,
@@ -2246,6 +2281,7 @@ export class ChatClient<
         })),
         forwardedProps: { ...mergedBody },
         ...(resumeItems ? { resume: resumeItems } : {}),
+        ...(byokHeaders ? { headers: byokHeaders } : {}),
       }
       this.devtoolsBridge.beginRun(runContext.runId, runContext.threadId)
       activeDevtoolsRunId = runContext.runId
@@ -2301,32 +2337,45 @@ export class ChatClient<
       // Finalize (idempotent — may already be done by RUN_FINISHED handler)
       this.processor.finalizeStream()
       streamCompletedSuccessfully = true
-    } catch (err) {
-      if (err instanceof Error) {
-        if (err.name === 'AbortError') {
-          if (activeDevtoolsRunId) {
-            this.devtoolsBridge.emitRunLifecycle(
-              'run:cancelled',
-              activeDevtoolsRunId,
-              'cancelled',
-            )
-            runTerminalEventEmitted = true
-          }
-          return false
+    } catch (err: unknown) {
+      const error = err instanceof Error ? err : new Error(String(err))
+      if (error.name === 'AbortError') {
+        if (activeDevtoolsRunId) {
+          this.devtoolsBridge.emitRunLifecycle(
+            'run:cancelled',
+            activeDevtoolsRunId,
+            'cancelled',
+          )
+          runTerminalEventEmitted = true
         }
-        if (generation === this.streamGeneration) {
-          this.reportStreamError(err)
-          if (activeDevtoolsRunId) {
-            this.devtoolsBridge.emitRunLifecycle(
-              'run:errored',
-              activeDevtoolsRunId,
-              'errored',
-              { error: err.message },
-            )
-            runTerminalEventEmitted = true
-          }
+        return false
+      }
+      if (error instanceof ByokMissingError) {
+        this.byok?.request(error.provider, 'missing')
+      }
+      if (error instanceof ByokBlockedError && error.reason === 'locked') {
+        this.byok?.request(error.provider, 'locked')
+      }
+      if (generation === this.streamGeneration) {
+        this.reportStreamError(error)
+        if (activeDevtoolsRunId) {
+          this.devtoolsBridge.emitRunLifecycle(
+            'run:errored',
+            activeDevtoolsRunId,
+            'errored',
+            { error: error.message },
+          )
+          runTerminalEventEmitted = true
         }
       }
+      if (
+        generation === this.streamGeneration &&
+        (error instanceof ByokMissingError ||
+          error instanceof ByokBlockedError ||
+          error instanceof ByokUnresolvedProviderError)
+      ) {
+        throw error
+      }
     } finally {
       // Only clean up if this is still the active stream.
       // A superseded stream (e.g. reload() started a new one) must not
@@ -2361,43 +2410,18 @@ export class ChatClient<
         // Drain any actions that were queued while the stream was in progress
         await this.drainPostStreamActions()
 
-        // Continue conversation if the stream ended with a tool result (server tool completed)
-        // but ONLY if the model indicated it wants to continue (finishReason !== 'stop').
-        // When finishReason is 'stop', the model is done — don't re-send.
         if (streamCompletedSuccessfully) {
-          const messages = this.processor.getMessages()
-          const lastPart = messages.at(-1)?.parts.at(-1)
-          const { finishReason } = this.processor.getState()
-
-          if (
-            lastPart?.type === 'tool-result' &&
-            finishReason !== 'stop' &&
-            this.shouldAutoSend()
-          ) {
-            try {
-              await this.checkForContinuation()
-            } catch (error) {
-              console.error('Failed to continue flow after tool result:', error)
-              // Continuation failed without starting a new stream — don't
-              // leave queued user messages stranded forever. (isLoading is
-              // already false in this finally block.)
-              await this.drainQueue()
-            }
-          } else {
-            if (this.status !== 'ready') {
-              // Terminal run, but onStreamEnd never fired: the processor had
-              // no assistant message to emit it for (e.g. a bare
-              // RUN_FINISHED{stop}, #421). The normal path already set
-              // 'ready', so this is a no-op.
-              this.setStatus('ready')
-            }
-            // Auto-send queued messages once the run fully settles. When a
-            // continuation runs instead (tool-result branch above), that
-            // continuation's own finally drains the queue. Skip if a drain
-            // loop is already walking the queue (avoids nested re-entry).
-            if (!this.messageQueueDraining) {
-              await this.drainQueue()
-            }
+          if (this.status !== 'ready') {
+            // Terminal run, but onStreamEnd never fired: the processor had
+            // no assistant message to emit it for (e.g. a bare
+            // RUN_FINISHED{stop}, #421). The normal path already set
+            // 'ready', so this is a no-op.
+            this.setStatus('ready')
+          }
+          // Auto-send queued messages once the run fully settles. Skip if a
+          // drain loop is already walking the queue (avoids nested re-entry).
+          if (!this.messageQueueDraining) {
+            await this.drainQueue()
           }
         } else {
           // Error/abort settle for the active generation: don't strand or
@@ -2979,6 +3003,12 @@ export class ChatClient<
     if (options.forwardedProps !== undefined) {
       this.forwardedPropsOption = options.forwardedProps
     }
+    if (options.byok !== undefined) {
+      this.byok = options.byok
+    }
+    if (options.byokProvider !== undefined) {
+      this.byokProvider = options.byokProvider
+    }
     if ('context' in options) {
       this.context = options.context
     }

package/src/connection-adapters.ts

@@ -6,6 +6,7 @@ import {
   uiMessagesToWire,
   withTanstackMetadata,
 } from '@tanstack/ai/client'
+import { ByokMissingError, isByokMissingBody } from '@tanstack/ai/byok'
 import {
   createResponseStreamTextDecoder,
   getResponseStreamReader,
@@ -498,12 +499,35 @@ async function* linesToNdjsonEvents(
   }
 }
 
-function assertResponseOk(response: Response): void {
-  if (!response.ok) {
-    throw new Error(
-      `HTTP error! status: ${response.status} ${response.statusText}`,
-    )
+async function assertResponseOk(response: Response): Promise<void> {
+  if (response.ok) return
+  if (response.status === 401) {
+    const body: unknown = await response
+      .clone()
+      .json()
+      .catch(() => null)
+    if (isByokMissingBody(body)) {
+      throw new ByokMissingError(body.error.provider)
+    }
+  }
+  throw new Error(
+    `HTTP error! status: ${response.status} ${response.statusText}`,
+  )
+}
+
+function errorFromXhrStatus(xhr: XMLHttpRequest): Error {
+  if (xhr.status === 401) {
+    let parsed: unknown = null
+    try {
+      parsed = JSON.parse(xhr.responseText)
+    } catch {
+      parsed = null
+    }
+    if (isByokMissingBody(parsed)) {
+      return new ByokMissingError(parsed.error.provider)
+    }
   }
+  return new Error(`XHR error! status: ${xhr.status} ${xhr.statusText}`)
 }
 
 /**
@@ -525,7 +549,7 @@ async function fetchThreadHydration(
     headers: { Accept: 'application/json', ...headers },
     credentials,
   })
-  assertResponseOk(response)
+  await assertResponseOk(response)
   const data = (await response.json()) as {
     messages?: Array<UIMessage>
     activeRun?: { runId?: unknown } | null
@@ -574,7 +598,7 @@ async function fetchGenerationHydration(
     headers: { Accept: 'application/json', ...headers },
     credentials,
   })
-  assertResponseOk(response)
+  await assertResponseOk(response)
   const raw: unknown = await response.json()
   // A 200 carrying `null` is a legitimate hydration miss — the server has no
   // record for this thread — and reading `.activeRun` off `null` would throw.
@@ -609,7 +633,7 @@ async function* responseToSSEEvents(
   abortSignal?: AbortSignal,
   fallbackIds?: { threadId?: string; runId?: string },
 ): AsyncGenerator<StreamEvent> {
-  assertResponseOk(response)
+  await assertResponseOk(response)
   const reader = getResponseStreamReader(response)
   yield* linesToSSEEvents(readStreamLines(reader, abortSignal), fallbackIds)
 }
@@ -619,7 +643,7 @@ async function* responseToNdjsonEvents(
   response: Response,
   abortSignal?: AbortSignal,
 ): AsyncGenerator<StreamEvent> {
-  assertResponseOk(response)
+  await assertResponseOk(response)
   const reader = getResponseStreamReader(response)
   yield* linesToNdjsonEvents(readStreamLines(reader, abortSignal))
 }
@@ -809,6 +833,8 @@ export interface RunAgentInputContext {
   }>
   /** Arbitrary user-controlled passthrough data. */
   forwardedProps?: Record<string, unknown>
+  /** Extra request headers for this run (e.g. BYOK keys). POST only. */
+  headers?: Record<string, string>
 }
 
 export interface ConnectConnectionAdapter {
@@ -1272,6 +1298,7 @@ export function fetchServerSentEvents(
       const requestHeaders: Record<string, string> = {
         'Content-Type': 'application/json',
         ...mergeHeaders(resolvedOptions.headers),
+        ...mergeHeaders(runContext?.headers),
         ...runIdHeader(runContext?.runId),
       }
 
@@ -1443,6 +1470,7 @@ export function fetchHttpStream(
       const requestHeaders: Record<string, string> = {
         'Content-Type': 'application/json',
         ...mergeHeaders(resolvedOptions.headers),
+        ...mergeHeaders(runContext?.headers),
         ...runIdHeader(runContext?.runId),
       }
 
@@ -1596,7 +1624,7 @@ function readXhrLines(
 
   const enqueueDelta = () => {
     if (xhr.status !== 0 && (xhr.status < 200 || xhr.status >= 300)) {
-      error = new Error(`XHR error! status: ${xhr.status} ${xhr.statusText}`)
+      error = errorFromXhrStatus(xhr)
       done = true
       return
     }
@@ -1625,7 +1653,7 @@ function readXhrLines(
     // is an error, but status 0 here is not — treat the trailing buffer as a
     // truncation check instead of fabricating a bogus "status: 0" error.
     if (xhr.status !== 0 && (xhr.status < 200 || xhr.status >= 300)) {
-      error = new Error(`XHR error! status: ${xhr.status} ${xhr.statusText}`)
+      error = errorFromXhrStatus(xhr)
     } else if (buffer.trim() && !aborted) {
       error = new StreamTruncatedError()
     }
@@ -1721,6 +1749,7 @@ function createConfiguredXhrRequest(
   const requestHeaders: Record<string, string> = {
     'Content-Type': 'application/json',
     ...mergeHeaders(options.headers),
+    ...mergeHeaders(method === 'POST' ? runContext?.headers : undefined),
     // Client-chosen run id for durability (POST only; the GET join carries it
     // in the query instead).
     ...(method === 'POST' ? runIdHeader(runContext?.runId) : {}),
@@ -2494,7 +2523,7 @@ export function fetcherToConnectionAdapter(
             ? { resume: runContext.resume }
             : {}),
         },
-        { signal: abortSignal },
+        { signal: abortSignal, headers: runContext.headers },
       )
       if (result instanceof Response) {
         yield* responseToSSEChunks(result, abortSignal)

package/src/generation-client.ts

@@ -1,3 +1,8 @@
+import { ByokBlockedError, ByokMissingError } from '@tanstack/ai/byok'
+import {
+  prepareResolvedByokHeaders,
+  resolveByokProviderId,
+} from './byok/resolve'
 import {
   GENERATION_EVENTS,
   GENERATION_STREAM_TRUNCATED_MESSAGE,
@@ -12,6 +17,7 @@ import { createNoOpGenerationDevtoolsBridge } from './devtools-noop'
 import { parseSSEResponse } from './sse-parser'
 import { restoreInboundChunk } from '@tanstack/ai/client'
 import type { StreamChunk } from '@tanstack/ai/client'
+import type { ByokClient } from './byok'
 import type {
   ConnectConnectionAdapter,
   GenerationHydrationResult,
@@ -120,6 +126,8 @@ export class GenerationClient<
   // the client hydrates the last generation for `threadId` from the server.
   private readonly serverDriven: boolean = false
   private body: Record<string, any>
+  private byok: ByokClient | undefined
+  private byokProvider: (() => string | undefined) | undefined
   private result: TOutput | null = null
   private input: TInput | null = null
   private progress: AIDevtoolsGenerationProgress | null = null
@@ -156,6 +164,8 @@ export class GenerationClient<
     this.hydrateGenerationHandler = options.hydrateGeneration
     this.joinRunHandler = options.joinRun
     this.body = options.body ?? {}
+    this.byok = options.byok
+    this.byokProvider = options.byokProvider
     // `persistence` is `false`/omitted (ephemeral) or `true` (server-driven:
     // hydrate the last generation for `threadId` from the server on mount).
     this.serverDriven = options.persistence === true
@@ -265,9 +275,21 @@ export class GenerationClient<
     const { signal } = abortController
 
     try {
+      let headers: Record<string, string> | undefined
+      if (this.byok) {
+        const provider = resolveByokProviderId(
+          this.byokProvider,
+          this.body.provider,
+        )
+        headers = await prepareResolvedByokHeaders(this.byok, provider)
+      }
+
       if (this.fetcher) {
         // Direct fetch path
-        const result = await this.fetcher(input, { signal })
+        const result = await this.fetcher(
+          input,
+          headers === undefined ? { signal } : { signal, headers },
+        )
         if (signal.aborted) return
         if (result instanceof Response) {
           // Server function returned SSE Response — parse stream
@@ -289,7 +311,7 @@ export class GenerationClient<
           [],
           mergedData,
           signal,
-          this.createRunContext(runId),
+          this.createRunContext(runId, headers),
         )
         await this.processStream(stream, runId, signal)
       } else {
@@ -312,6 +334,12 @@ export class GenerationClient<
     } catch (err: unknown) {
       if (signal.aborted) return
       const error = err instanceof Error ? err : new Error(String(err))
+      if (error instanceof ByokMissingError) {
+        this.byok?.request(error.provider, 'missing')
+      }
+      if (error instanceof ByokBlockedError && error.reason === 'locked') {
+        this.byok?.request(error.provider, 'locked')
+      }
       this.setError(error)
       this.setStatus('error')
       this.recordResumeSnapshotError(error)
@@ -462,13 +490,25 @@ export class GenerationClient<
     options: Partial<
       Pick<
         GenerationClientOptions<TInput, TResult, TOutput>,
-        'body' | 'onResult' | 'onError' | 'onProgress' | 'onChunk'
+        | 'body'
+        | 'byok'
+        | 'byokProvider'
+        | 'onResult'
+        | 'onError'
+        | 'onProgress'
+        | 'onChunk'
       >
     >,
   ): void {
     if (options.body !== undefined) {
       this.body = options.body ?? {}
     }
+    if (options.byok !== undefined) {
+      this.byok = options.byok
+    }
+    if (options.byokProvider !== undefined) {
+      this.byokProvider = options.byokProvider
+    }
     if (options.onResult !== undefined) {
       this.callbacksRef.onResult = options.onResult
     }
@@ -645,10 +685,14 @@ export class GenerationClient<
     return `${prefix}-${Date.now()}-${Math.random().toString(36).substring(7)}`
   }
 
-  private createRunContext(runId: string): RunAgentInputContext {
+  private createRunContext(
+    runId: string,
+    headers?: Record<string, string>,
+  ): RunAgentInputContext {
     return {
       threadId: this.threadId,
       runId,
+      ...(headers ? { headers } : {}),
     }
   }
 

package/src/generation-types.ts

@@ -5,6 +5,7 @@ import type {
   StreamChunk,
 } from '@tanstack/ai/client'
 import type { TokenUsage, TranscriptionResponseFormat } from '@tanstack/ai'
+import type { ByokClient } from './byok'
 import type { ConnectConnectionAdapter } from './connection-adapters'
 import type { AIDevtoolsClientMetadata } from './devtools'
 import type {
@@ -271,6 +272,8 @@ export const GENERATION_EVENTS = {
 export interface GenerationFetcherOptions {
   /** AbortSignal that is triggered when the user calls `stop()` */
   signal: AbortSignal
+  /** Extra request headers for this run (e.g. BYOK keys). */
+  headers?: Record<string, string>
 }
 
 /**
@@ -338,6 +341,19 @@ export interface GenerationClientOptions<_TInput, TResult, TOutput = TResult> {
   /** Additional body parameters to send with connect-based adapter requests */
   body?: Record<string, any>
 
+  /**
+   * Optional BYOK keyring. On each generate the client prepares the resolved
+   * provider and stamps `x-byok-*` request headers. Keys never go in the body.
+   */
+  byok?: ByokClient
+
+  /**
+   * Optional provider id for this generation. If it returns a provider slug,
+   * only that key is prepared and sent. Otherwise `body.provider` is used.
+   * If no slug resolves, generate throws instead of attaching every stored key.
+   */
+  byokProvider?: () => string | undefined
+
   /** Metadata used to register this generation hook with TanStack AI Devtools */
   devtools?: Partial<AIDevtoolsClientMetadata>
 

package/src/interrupt-manager.ts

@@ -102,6 +102,10 @@ interface TransactionToken {
   active: boolean
 }
 
+interface SubmissionOperation {
+  submission: InterruptManagerSubmission
+}
+
 interface RuntimeInterruptCheckpoint {
   status: InterruptItemStatus
   resolution?: RunAgentResumeItem
@@ -511,6 +515,7 @@ export class InterruptManager<
     resuming: false,
   })
   private activeTransaction: TransactionToken | undefined
+  private activeSubmissionOperation: SubmissionOperation | undefined
   private retrySubmission: InterruptManagerSubmission | undefined
   private resuming = false
   private tools: TTools | undefined
@@ -544,6 +549,7 @@ export class InterruptManager<
     hydration: InterruptManagerHydration,
     source: InterruptManagerChangeSource = 'live',
   ): void {
+    this.activeSubmissionOperation = undefined
     this.hydration = {
       threadId: hydration.threadId,
       interruptedRunId: hydration.interruptedRunId,
@@ -597,6 +603,7 @@ export class InterruptManager<
     preserveRootErrors?: boolean
     source?: InterruptManagerChangeSource
   }): void {
+    this.activeSubmissionOperation = undefined
     this.hydration = undefined
     this.items = []
     this.snapshot = Object.freeze([])
@@ -1528,25 +1535,32 @@ export class InterruptManager<
   }
 
   private submitBatch(submission: InterruptManagerSubmission): void {
+    // Track ownership so a superseded submission cannot mutate current state.
+    const operation = { submission }
+    this.activeSubmissionOperation = operation
     this.resuming = true
     this.retrySubmission = undefined
     for (const item of this.items) {
       if (isClientOwnedInterrupt(item)) item.status = 'submitting'
     }
     this.publish()
-    void this.performSubmission(submission)
+    void this.performSubmission(operation)
   }
 
   private async performSubmission(
-    submission: InterruptManagerSubmission,
+    operation: SubmissionOperation,
   ): Promise<void> {
     try {
-      await this.options.submit(submission)
+      await this.options.submit(operation.submission)
     } catch (error) {
-      this.handleSubmissionFailure(error, submission)
+      if (this.activeSubmissionOperation !== operation) return
+      this.handleSubmissionFailure(error, operation.submission)
     } finally {
-      this.resuming = false
-      this.publish()
+      if (this.activeSubmissionOperation === operation) {
+        this.activeSubmissionOperation = undefined
+        this.resuming = false
+        this.publish()
+      }
     }
   }
 

package/src/sse-parser.ts

@@ -1,3 +1,4 @@
+import { ByokMissingError, isByokMissingBody } from '@tanstack/ai/byok'
 import {
   createResponseStreamTextDecoder,
   getResponseStreamReader,
@@ -51,6 +52,15 @@ export async function* parseSSEResponse(
   abortSignal?: AbortSignal,
 ): AsyncGenerator<StreamChunk> {
   if (!response.ok) {
+    if (response.status === 401) {
+      const body: unknown = await response
+        .clone()
+        .json()
+        .catch(() => null)
+      if (isByokMissingBody(body)) {
+        throw new ByokMissingError(body.error.provider)
+      }
+    }
     throw new Error(
       `HTTP error! status: ${response.status} ${response.statusText}`,
     )