@vellumai/assistant 0.8.4 → 0.8.6

package/src/runtime/routes/inbound-stages/background-dispatch.ts

@@ -11,7 +11,11 @@ import type { ChannelId, InterfaceId } from "../../../channels/types.js";
 import { findGuardianForChannel } from "../../../contacts/contact-store.js";
 import type { ServerMessage } from "../../../daemon/message-protocol.js";
 import type { TrustContext } from "../../../daemon/trust-context.js";
-import { updateDeliveredSegmentCount } from "../../../memory/delivery-channels.js";
+import {
+  addSlackDmLiveDeliveredTextResponseIndex,
+  getSlackDmLiveDeliveredTextResponseIndexes,
+  updateDeliveredSegmentCount,
+} from "../../../memory/delivery-channels.js";
 import {
   linkMessage,
   storeReplyMessageId,
@@ -44,6 +48,10 @@ import type {
   MessageProcessor,
   SlackInboundMessageMetadata,
 } from "../../http-types.js";
+import {
+  createSlackDmTextDeliveryController,
+  isSlackDeliveryCallbackUrl,
+} from "../../slack-dm-text-delivery.js";
 import { resolveRoutingState } from "../../trust-context-resolver.js";
 import { deliverReplyViaCallback } from "../channel-delivery-routes.js";
 import { deliverGeneratedApprovalPrompt } from "../guardian-approval-prompt.js";
@@ -228,6 +236,21 @@ export function processChannelMessageInBackground(
             }
           : undefined;
       let replyMessageId: string | undefined;
+      const slackDmTextDelivery = createSlackDmTextDeliveryController({
+        sourceChannel,
+        chatType,
+        replyCallbackUrl,
+        chatId: externalChatId,
+        assistantId,
+        deliveredTextResponseIndexes:
+          getSlackDmLiveDeliveredTextResponseIndexes(eventId),
+        onTextResponseDelivered: (responseIndex, reason) => {
+          addSlackDmLiveDeliveredTextResponseIndex(eventId, responseIndex);
+          if (reason === "before_tool") {
+            slackThinkingStatus?.refreshAfterReply();
+          }
+        },
+      });
       const observeAgentEvent = (msg: ServerMessage): void => {
         if (
           msg.type === "message_complete" &&
@@ -236,6 +259,7 @@ export function processChannelMessageInBackground(
         ) {
           replyMessageId = msg.messageId;
         }
+        slackDmTextDelivery?.observeEvent(msg);
         slackThinkingStatus?.observeEvent(msg);
       };
 
@@ -297,12 +321,21 @@ export function processChannelMessageInBackground(
           { err, conversationId },
           "Background channel message processing failed",
         );
+        if (slackDmTextDelivery) {
+          await slackDmTextDelivery.waitForPendingDeliveries();
+        }
         recordProcessingFailure(eventId, err);
         return;
       }
 
       if (replyCallbackUrl) {
         try {
+          if (slackDmTextDelivery) {
+            await slackDmTextDelivery.waitForPendingDeliveries();
+          }
+          const liveDeliveryResumeOptions =
+            slackDmTextDelivery?.getFinalDeliveryResumeOptions(replyMessageId);
+
           await deliverReplyViaCallback(
             conversationId,
             externalChatId,
@@ -311,6 +344,7 @@ export function processChannelMessageInBackground(
             {
               messageId: replyMessageId,
               sinceMessageId: userMessageId,
+              ...liveDeliveryResumeOptions,
               onSegmentDelivered: (count) =>
                 updateDeliveredSegmentCount(eventId, count),
             },
@@ -395,11 +429,13 @@ function startTelegramTypingHeartbeat(
 
 type SlackThinkingStatusController = {
   observeEvent: (msg: ServerMessage) => void;
+  refreshAfterReply: () => void;
   stop: () => void;
 };
 
 type SlackThinkingStatusHandle = {
   updateLoadingMessages: (loadingMessages?: string[]) => void;
+  refresh: (loadingMessages?: string[]) => void;
   clear: () => void;
 };
 
@@ -440,12 +476,9 @@ function shouldEmitSlackThinkingStatus(
   sourceChannel: ChannelId,
   replyCallbackUrl?: string,
 ): boolean {
-  if (sourceChannel !== "slack" || !replyCallbackUrl) return false;
-  try {
-    return new URL(replyCallbackUrl).pathname.endsWith("/deliver/slack");
-  } catch {
-    return replyCallbackUrl.endsWith("/deliver/slack");
-  }
+  return (
+    sourceChannel === "slack" && isSlackDeliveryCallbackUrl(replyCallbackUrl)
+  );
 }
 
 export function shouldStartSlackThinkingStatusImmediately(params: {
@@ -547,6 +580,13 @@ function createSlackThinkingStatusController(params: {
         start();
       }
     },
+    refreshAfterReply() {
+      if (stopped || !slackThinkingStatus) return;
+      // Slack clears assistant thread status when the app sends a reply, so
+      // live pre-tool replies need to reassert it while work continues.
+      slackThinkingStatus.refresh(currentLoadingMessages);
+      lastSentLoadingMessageKey = getLoadingMessagesKey(currentLoadingMessages);
+    },
     stop() {
       stopped = true;
       slackThinkingStatus?.clear();
@@ -627,7 +667,9 @@ function getTaskProgressLoadingMessage(
 
   const activeStep = progress.steps[activeStepIndex]!;
   return [
-    `In progress (${activeStepIndex + 1}/${progress.steps.length}): ${activeStep.label}`,
+    `In progress (${activeStepIndex + 1}/${progress.steps.length}): ${
+      activeStep.label
+    }`,
   ];
 }
 
@@ -656,6 +698,7 @@ function setSlackThinkingStatus(
     if (!messageTs) {
       return {
         updateLoadingMessages: () => {},
+        refresh: () => {},
         clear: () => {},
       };
     }
@@ -697,6 +740,7 @@ function setSlackThinkingStatus(
 
     return {
       updateLoadingMessages: () => {},
+      refresh: () => {},
       clear: clearReaction,
     };
   }
@@ -739,6 +783,10 @@ function setSlackThinkingStatus(
     );
   };
 
+  const refresh = (nextLoadingMessages?: string[]): void => {
+    updateLoadingMessages(nextLoadingMessages);
+  };
+
   const clearStatus = (): void => {
     if (cleared) return;
     cleared = true;
@@ -766,6 +814,7 @@ function setSlackThinkingStatus(
 
   return {
     updateLoadingMessages,
+    refresh,
     clear: clearStatus,
   };
 }

package/src/runtime/routes/index.ts

@@ -18,10 +18,12 @@ import { ROUTES as AUDIT_ROUTES } from "./audit-routes.js";
 import { ROUTES as AUTH_ROUTES } from "./auth-routes.js";
 import { ROUTES as AVATAR_ROUTES } from "./avatar-routes.js";
 import { ROUTES as BACKGROUND_TOOL_ROUTES } from "./background-tool-routes.js";
+import { ROUTES as BACKGROUND_WAKE_ROUTES } from "./background-wake-routes.js";
 import { ROUTES as BACKUP_ROUTES } from "./backup-routes.js";
 import { ROUTES as BOOKMARK_ROUTES } from "./bookmark-routes.js";
 import { ROUTES as BRAIN_GRAPH_ROUTES } from "./brain-graph-routes.js";
 import { ROUTES as BROWSER_ROUTES } from "./browser-routes.js";
+import { ROUTES as BROWSER_TABS_ROUTES } from "./browser-tabs-routes.js";
 import { ROUTES as BTW_ROUTES } from "./btw-routes.js";
 import { ROUTES as CACHE_ROUTES } from "./cache-routes.js";
 import { ROUTES as CALL_ROUTES } from "./call-routes.js";
@@ -38,6 +40,7 @@ import { ROUTES as CONTENT_SOURCE_ROUTES } from "./content-source-routes.js";
 import { ROUTES as CONVERSATION_ANALYSIS_ROUTES } from "./conversation-analysis-routes.js";
 import { ROUTES as CONVERSATION_ATTENTION_ROUTES } from "./conversation-attention-routes.js";
 import { ROUTES as CONVERSATION_CLI_ROUTES } from "./conversation-cli-routes.js";
+import { ROUTES as CONVERSATION_COMPACTION_ROUTES } from "./conversation-compaction-routes.js";
 import { ROUTES as CONVERSATION_LIST_ROUTES } from "./conversation-list-routes.js";
 import { ROUTES as CONVERSATION_MANAGEMENT_ROUTES } from "./conversation-management-routes.js";
 import { ROUTES as CONVERSATION_QUERY_ROUTES } from "./conversation-query-routes.js";
@@ -88,6 +91,7 @@ import { ROUTES as LOG_EXPORT_ROUTES } from "./log-export-routes.js";
 import { ROUTES as MCP_AUTH_ROUTES } from "./mcp-auth-routes.js";
 import { ROUTES as MEMORY_ITEM_ROUTES } from "./memory-item-routes.js";
 import { ROUTES as MEMORY_V2_ROUTES } from "./memory-v2-routes.js";
+import { ROUTES as MEMORY_V3_ROUTES } from "./memory-v3-routes.js";
 import { ROUTES as MIGRATION_ROLLBACK_ROUTES } from "./migration-rollback-routes.js";
 import { ROUTES as MIGRATION_ROUTES } from "./migration-routes.js";
 import { ROUTES as NOTIFICATION_ROUTES } from "./notification-routes.js";
@@ -98,6 +102,7 @@ import { ROUTES as OAUTH_LIFECYCLE_ROUTES } from "./oauth-lifecycle-routes.js";
 import { ROUTES as OAUTH_PROVIDERS_ROUTES } from "./oauth-providers.js";
 import { ROUTES as PLATFORM_ROUTES } from "./platform-routes.js";
 import { ROUTES as PLAYGROUND_ROUTES } from "./playground/index.js";
+import { ROUTES as PLUGINS_ROUTES } from "./plugins-routes.js";
 import { ROUTES as PROFILER_ROUTES } from "./profiler-routes.js";
 import { ROUTES as PS_ROUTES } from "./ps-routes.js";
 import { ROUTES as PUBLISH_ROUTES } from "./publish-routes.js";
@@ -145,6 +150,7 @@ export const ROUTES: RouteDefinition[] = [
   ...AUDIT_ROUTES,
   ...AUTH_ROUTES,
   ...AVATAR_ROUTES,
+  ...BACKGROUND_WAKE_ROUTES,
   ...BACKGROUND_TOOL_ROUTES,
   ...BACKUP_ROUTES,
   ...BOOKMARK_ROUTES,
@@ -156,6 +162,7 @@ export const ROUTES: RouteDefinition[] = [
   ...CHANNEL_AVAILABILITY_ROUTES,
   ...CHANNEL_READINESS_ROUTES,
   ...BROWSER_ROUTES,
+  ...BROWSER_TABS_ROUTES,
   ...BTW_ROUTES,
   ...BRAIN_GRAPH_ROUTES,
   ...CLIENT_ROUTES,
@@ -173,6 +180,7 @@ export const ROUTES: RouteDefinition[] = [
   ...CREDENTIAL_PROMPT_ROUTES,
   ...CREDENTIAL_ROUTES,
   ...DEFER_ROUTES,
+  ...CONVERSATION_COMPACTION_ROUTES,
   ...CONVERSATION_QUERY_ROUTES,
   ...CONVERSATION_STARTER_ROUTES,
   ...DEBUG_BASH_ROUTES,
@@ -211,6 +219,7 @@ export const ROUTES: RouteDefinition[] = [
   ...LLM_CALL_SITES_ROUTES,
   ...MEMORY_ITEM_ROUTES,
   ...MEMORY_V2_ROUTES,
+  ...MEMORY_V3_ROUTES,
   ...MIGRATION_ROLLBACK_ROUTES,
   ...MIGRATION_ROUTES,
   ...NOTIFICATION_ROUTES,
@@ -220,6 +229,7 @@ export const ROUTES: RouteDefinition[] = [
   ...OAUTH_PROVIDERS_ROUTES,
   ...PLATFORM_ROUTES,
   ...PLAYGROUND_ROUTES,
+  ...PLUGINS_ROUTES,
   ...PROFILER_ROUTES,
   ...PS_ROUTES,
   ...PUBLISH_ROUTES,

package/src/runtime/routes/inference-profile-session-handler.ts

@@ -67,11 +67,13 @@ export async function setInferenceProfileSession({
   profile,
   ttlSeconds,
   sessionId: callerSessionId,
+  originClientId,
 }: {
   conversationId: string;
   profile: string | null;
   ttlSeconds?: number | null;
   sessionId?: string;
+  originClientId?: string;
 }): Promise<InferenceProfileSessionResult> {
   const resolvedId = resolveConversationId(conversationId) ?? conversationId;
   const conversation = getConversation(resolvedId);
@@ -116,12 +118,15 @@ export async function setInferenceProfileSession({
       };
     }
     setConversationInferenceProfileSession(resolvedId, null, null, null);
-    publishConversationInferenceProfileChanged({
-      conversationId: resolvedId,
-      profile: null,
-      sessionId: null,
-      expiresAt: null,
-    });
+    publishConversationInferenceProfileChanged(
+      {
+        conversationId: resolvedId,
+        profile: null,
+        sessionId: null,
+        expiresAt: null,
+      },
+      originClientId,
+    );
     return {
       conversationId: resolvedId,
       profile: null,
@@ -184,12 +189,15 @@ export async function setInferenceProfileSession({
     newExpiresAt ?? null,
   );
 
-  publishConversationInferenceProfileChanged({
-    conversationId: resolvedId,
-    profile,
-    sessionId: newSessionId ?? null,
-    expiresAt: newExpiresAt ?? null,
-  });
+  publishConversationInferenceProfileChanged(
+    {
+      conversationId: resolvedId,
+      profile,
+      sessionId: newSessionId ?? null,
+      expiresAt: newExpiresAt ?? null,
+    },
+    originClientId,
+  );
 
   return {
     conversationId: resolvedId,
@@ -218,6 +226,7 @@ export async function setInferenceProfileSession({
  */
 export async function closeInferenceProfileSession(
   conversationId: string,
+  originClientId?: string,
 ): Promise<{
   conversationId: string;
   closed: { profile: string | null; sessionId: string | null } | null;
@@ -241,6 +250,7 @@ export async function closeInferenceProfileSession(
   const result = await setInferenceProfileSession({
     conversationId: resolvedId,
     profile: null,
+    originClientId,
   });
   return {
     conversationId: result.conversationId,

package/src/runtime/routes/inference-profile-session-routes.ts

@@ -22,6 +22,7 @@ import type { RouteDefinition, RouteHandlerArgs } from "./types.js";
 
 async function handleOpenInferenceProfileSession({
   body = {},
+  headers,
 }: RouteHandlerArgs) {
   if (body.profile == null || typeof body.profile !== "string") {
     throw new BadRequestError("profile must be a non-empty string");
@@ -31,13 +32,18 @@ async function handleOpenInferenceProfileSession({
     profile: body.profile,
     ttlSeconds: body.ttlSeconds as number | null | undefined,
     sessionId: body.sessionId as string | undefined,
+    originClientId: headers?.["x-vellum-client-id"]?.trim() || undefined,
   });
 }
 
 async function handleCloseInferenceProfileSession({
   body = {},
+  headers,
 }: RouteHandlerArgs) {
-  return closeInferenceProfileSession(body.conversationId as string);
+  return closeInferenceProfileSession(
+    body.conversationId as string,
+    headers?.["x-vellum-client-id"]?.trim() || undefined,
+  );
 }
 
 function handleListInferenceProfileSessions({

package/src/runtime/routes/inference-provider-connection-routes.ts

@@ -4,7 +4,7 @@
  * GET    /v1/inference/provider-connections          — list all connections (optional ?provider= filter)
  * GET    /v1/inference/provider-connections/:name    — single connection by name
  * POST   /v1/inference/provider-connections          — create a new connection
- * PATCH  /v1/inference/provider-connections/:name    — update auth/label/status (cannot rename or change provider; auth is locked to platform for managed connections)
+ * PATCH  /v1/inference/provider-connections/:name    — update auth/label (cannot rename or change provider; auth is locked to platform for managed connections)
  * DELETE /v1/inference/provider-connections/:name    — delete (rejects if profiles or call sites reference it; rejects outright for managed connections)
  */
 
@@ -18,7 +18,6 @@ import {
   type ConnectionModel,
   ConnectionModelSchema,
   ConnectionProviderSchema,
-  ConnectionStatusSchema,
   ProviderConnectionSchema,
   VALID_CONNECTION_PROVIDERS,
 } from "../../providers/inference/auth.js";
@@ -223,14 +222,6 @@ async function handleCreateConnection({ body = {} }: RouteHandlerArgs) {
     throw new BadRequestError(`Invalid auth: ${authResult.error.message}`);
   }
 
-  const statusResult =
-    body.status !== undefined
-      ? ConnectionStatusSchema.safeParse(body.status)
-      : null;
-  if (statusResult && !statusResult.success) {
-    throw new BadRequestError(`Invalid status: must be "active" or "disabled"`);
-  }
-
   const labelRaw = body.label;
   if (
     labelRaw !== undefined &&
@@ -248,7 +239,6 @@ async function handleCreateConnection({ body = {} }: RouteHandlerArgs) {
     name,
     provider: providerResult.data,
     auth: authResult.data,
-    ...(statusResult ? { status: statusResult.data } : {}),
     ...(labelRaw !== undefined ? { label: labelRaw as string | null } : {}),
     ...customFields,
   });
@@ -302,14 +292,6 @@ async function handleUpdateConnection({
     throw new BadRequestError(`Invalid auth: ${authResult.error.message}`);
   }
 
-  const statusResult =
-    body.status !== undefined
-      ? ConnectionStatusSchema.safeParse(body.status)
-      : null;
-  if (statusResult && !statusResult.success) {
-    throw new BadRequestError(`Invalid status: must be "active" or "disabled"`);
-  }
-
   const labelRaw = body.label;
   if (
     labelRaw !== undefined &&
@@ -323,8 +305,8 @@ async function handleUpdateConnection({
 
   // Managed connections: lock auth to `{type:"platform"}`. The boot upsert in
   // `seedCanonicalConnections` would revert any other value on next restart;
-  // reject the write here so the surprise loop never happens. Label and status
-  // remain user-editable (the boot upsert leaves those alone).
+  // reject the write here so the surprise loop never happens. Label remains
+  // user-editable (the boot upsert leaves it alone).
   if (
     MANAGED_CONNECTION_NAMES.has(name) &&
     authResult.data.type !== "platform"
@@ -338,7 +320,6 @@ async function handleUpdateConnection({
 
   const result = updateConnection(getDb(), name, {
     auth: authResult.data,
-    ...(statusResult ? { status: statusResult.data } : {}),
     ...(labelRaw !== undefined ? { label: labelRaw as string | null } : {}),
     ...customFields,
   });
@@ -387,7 +368,7 @@ function handleDeleteConnection({ pathParams = {} }: RouteHandlerArgs) {
   // re-overlaid by `seed-inference-profiles.ts` on boot).
   if (MANAGED_CONNECTION_NAMES.has(name)) {
     throw new BadRequestError(
-      `Cannot delete managed connection "${name}". This is a Vellum-managed connection — disable it via PATCH status="disabled" if you want to opt out of platform inference.`,
+      `Cannot delete managed connection "${name}". This is a Vellum-managed connection that is re-seeded on every startup.`,
     );
   }
 
@@ -485,7 +466,6 @@ export const ROUTES: RouteDefinition[] = [
       provider: ConnectionProviderSchema,
       auth: AuthSchema,
       label: z.string().min(1).optional(),
-      status: ConnectionStatusSchema.optional(),
       base_url: z.string().url().nullable().optional(),
       models: z.array(ConnectionModelSchema).nullable().optional(),
     }),
@@ -504,12 +484,11 @@ export const ROUTES: RouteDefinition[] = [
     policyKey: "inference/provider-connections/detail",
     summary: "Update a provider connection",
     description:
-      "Update an existing connection. Cannot rename or change the provider. For managed connections (anthropic-managed, openai-managed, gemini-managed) the auth is locked to platform; label and status remain editable.",
+      "Update an existing connection. Cannot rename or change the provider. For managed connections (anthropic-managed, openai-managed, gemini-managed) the auth is locked to platform; label remains editable.",
     tags: ["inference"],
     pathParams: [{ name: "name", description: "Connection name" }],
     requestBody: z.object({
       auth: AuthSchema,
-      status: ConnectionStatusSchema.optional(),
       label: z.string().min(1).nullable().optional(),
       base_url: z.string().url().nullable().optional(),
       models: z.array(ConnectionModelSchema).nullable().optional(),

package/src/runtime/routes/integrations/vercel.ts

@@ -9,6 +9,8 @@
  * ("set" or "delete") rather than using HTTP verbs directly.
  */
 
+import { z } from "zod";
+
 import {
   deleteVercelConfig,
   getVercelConfig,
@@ -17,6 +19,12 @@ import {
 import { BadRequestError } from "../errors.js";
 import type { RouteDefinition, RouteHandlerArgs } from "../types.js";
 
+const vercelConfigResponseSchema = z.object({
+  hasToken: z.boolean(),
+  success: z.boolean(),
+  error: z.string().optional(),
+});
+
 // ---------------------------------------------------------------------------
 // Handlers
 // ---------------------------------------------------------------------------
@@ -66,6 +74,7 @@ export const ROUTES: RouteDefinition[] = [
     description: "Check if a Vercel API token is stored.",
     tags: ["integrations"],
     requirePolicyEnforcement: true,
+    responseBody: vercelConfigResponseSchema,
     handler: () => handleGetVercelConfig(),
   },
   {
@@ -77,6 +86,11 @@ export const ROUTES: RouteDefinition[] = [
       "Set or delete the Vercel API token. Action is determined by the body action field.",
     tags: ["integrations"],
     requirePolicyEnforcement: true,
+    requestBody: z.object({
+      action: z.enum(["get", "set", "delete"]).optional(),
+      apiToken: z.string().optional(),
+    }),
+    responseBody: vercelConfigResponseSchema,
     handler: handlePostVercelConfig,
   },
   {
@@ -87,6 +101,7 @@ export const ROUTES: RouteDefinition[] = [
     description: "Delete the stored Vercel API token.",
     tags: ["integrations"],
     requirePolicyEnforcement: true,
+    responseBody: vercelConfigResponseSchema,
     handler: () => handleDeleteVercelConfig(),
   },
 ];

package/src/runtime/routes/llm-call-sites-routes.ts

@@ -1,6 +1,9 @@
 import { resolveDefaultProfileKey } from "../../config/llm-resolver.js";
 import { loadConfig } from "../../config/loader.js";
-import { CALL_SITE_CATALOG, CALL_SITE_DOMAINS } from "../../config/schemas/call-site-catalog.js";
+import {
+  CALL_SITE_CATALOG,
+  CALL_SITE_DOMAINS,
+} from "../../config/schemas/call-site-catalog.js";
 import type { LLMCallSite } from "../../config/schemas/llm.js";
 import type { RouteDefinition } from "./types.js";
 
@@ -10,14 +13,28 @@ async function handleGetCallSites() {
     domains: CALL_SITE_DOMAINS,
     callSites: CALL_SITE_CATALOG.map((entry) => ({
       ...entry,
-      defaultProfile: resolveDefaultProfileKey(
-        entry.id as LLMCallSite,
-        llm,
-      ),
+      defaultProfile: resolveDefaultProfileKey(entry.id as LLMCallSite, llm),
     })),
   };
 }
 
+export interface LlmProfilesListResult {
+  /** Sorted list of profile names defined in `llm.profiles`. */
+  profiles: string[];
+  /** The workspace-wide active profile name, if one is set. */
+  activeProfile: string | null;
+}
+
+async function handleListProfiles(): Promise<LlmProfilesListResult> {
+  const { llm } = loadConfig();
+  const profiles = llm?.profiles ?? {};
+  return {
+    profiles: Object.keys(profiles).sort(),
+    activeProfile:
+      typeof llm?.activeProfile === "string" ? llm.activeProfile : null,
+  };
+}
+
 export const ROUTES: RouteDefinition[] = [
   {
     operationId: "llm_call_sites_list",
@@ -29,4 +46,14 @@ export const ROUTES: RouteDefinition[] = [
       "Returns the full catalog of LLM call sites with display names, descriptions, and domain groupings. Used by clients to render the per-call-site override settings UI.",
     tags: ["config"],
   },
+  {
+    operationId: "llm_profiles_list",
+    method: "GET",
+    endpoint: "config/llm/profiles",
+    handler: handleListProfiles,
+    summary: "List defined LLM profiles",
+    description:
+      "Returns the sorted list of profile names defined in `llm.profiles` plus the workspace-wide active profile. Used to populate per-call profile dropdowns (e.g. memory router playground) without requiring the caller to type profile names.",
+    tags: ["config"],
+  },
 ];

package/src/runtime/routes/llm-context-normalization.ts

@@ -1057,8 +1057,13 @@ function extractOpenAiResponsesRequestToolNames(tools: unknown): string[] {
       if (asString(tool.type) === "function" && asString(tool.name)) {
         return asString(tool.name);
       }
-      // Native web search tool: { type: "web_search_preview" }
-      if (asString(tool.type) === "web_search_preview") {
+      // Native web search tools:
+      // - OpenAI API: { type: "web_search_preview" }
+      // - Codex subscription endpoint: { type: "web_search" }
+      if (
+        asString(tool.type) === "web_search_preview" ||
+        asString(tool.type) === "web_search"
+      ) {
         return "web_search";
       }
       return undefined;

package/src/runtime/routes/memory-item-routes.ts

@@ -44,7 +44,10 @@ import type {
   MemoryType,
   NewNode,
 } from "../../memory/graph/types.js";
-import { enqueueMemoryJob } from "../../memory/jobs-store.js";
+import {
+  enqueueMemoryJob,
+  isMemoryEnabled,
+} from "../../memory/jobs-store.js";
 import { withQdrantBreaker } from "../../memory/qdrant-circuit-breaker.js";
 import { getQdrantClient } from "../../memory/qdrant-client.js";
 import { memoryGraphNodes } from "../../memory/schema.js";
@@ -525,7 +528,9 @@ async function handleCreateMemoryItem(body: Record<string, unknown>) {
   };
 
   const created = createNode(newNode);
-  enqueueMemoryJob("embed_graph_node", { nodeId: created.id });
+  if (isMemoryEnabled()) {
+    enqueueMemoryJob("embed_graph_node", { nodeId: created.id });
+  }
 
   return { item: nodeToPayload(created) };
 }
@@ -619,7 +624,7 @@ async function handleUpdateMemoryItem(
 
   updateNode(id, changes);
 
-  if (contentChanged) {
+  if (contentChanged && isMemoryEnabled()) {
     enqueueMemoryJob("embed_graph_node", { nodeId: id });
   }