@agent-native/core 0.119.3 → 0.119.5

package/corpus/core/src/agent/thread-data-builder.ts

@@ -35,6 +35,7 @@ interface BuildAssistantMessageOptions {
    * overwriting the others.
    */
   turnId?: string;
+  runDurationMs?: number;
 }
 
 type AssistantMessage = NonNullable<ReturnType<typeof buildAssistantMessage>>;
@@ -43,6 +44,8 @@ type UserMessage = ReturnType<typeof buildUserMessage>;
 const INTERRUPTED_TOOL_RESULT =
   "Interrupted before this tool returned a result.";
 
+export const ASSISTANT_RUN_DURATION_METADATA_KEY = "agentNativeRunDurationMs";
+
 const MAX_STORED_ATTACHMENT_CHARS = 60_000;
 /**
  * When no file-upload provider is configured we fall back to storing base64
@@ -267,6 +270,13 @@ export function buildAssistantMessage(
   const custom: Record<string, unknown> = {};
   if (options.turnId) custom.turnId = options.turnId;
   if (runId) custom.foldedRunIds = [runId];
+  if (
+    typeof options.runDurationMs === "number" &&
+    Number.isFinite(options.runDurationMs) &&
+    options.runDurationMs >= 0
+  ) {
+    custom[ASSISTANT_RUN_DURATION_METADATA_KEY] = options.runDurationMs;
+  }
   if (continued) custom.continued = true;
   if (runError) {
     custom.runError = {
@@ -462,6 +472,41 @@ function messagesMatch(a: any, b: any): boolean {
   return messageIdentityKeys(a).some((key) => bKeys.has(key));
 }
 
+function preserveAssistantRunDuration(chosenEntry: any, otherEntry: any): any {
+  const chosen = getStoredMessage(chosenEntry);
+  const other = getStoredMessage(otherEntry);
+  if (chosen?.role !== "assistant" || other?.role !== "assistant") {
+    return chosenEntry;
+  }
+
+  const chosenCustom =
+    chosen.metadata?.custom && typeof chosen.metadata.custom === "object"
+      ? (chosen.metadata.custom as Record<string, unknown>)
+      : {};
+  if (assistantRunDurationMs(chosenCustom) != null) return chosenEntry;
+
+  const otherCustom =
+    other.metadata?.custom && typeof other.metadata.custom === "object"
+      ? (other.metadata.custom as Record<string, unknown>)
+      : {};
+  const durationMs = assistantRunDurationMs(otherCustom);
+  if (durationMs == null) return chosenEntry;
+
+  const nextMessage = {
+    ...chosen,
+    metadata: {
+      ...chosen.metadata,
+      custom: {
+        ...chosenCustom,
+        [ASSISTANT_RUN_DURATION_METADATA_KEY]: durationMs,
+      },
+    },
+  };
+  return chosenEntry?.message === undefined
+    ? nextMessage
+    : { ...chosenEntry, message: nextMessage };
+}
+
 function chooseMergedMessageEntry(existingEntry: any, incomingEntry: any): any {
   const existing = getStoredMessage(existingEntry);
   const incoming = getStoredMessage(incomingEntry);
@@ -479,12 +524,19 @@ function chooseMergedMessageEntry(existingEntry: any, incomingEntry: any): any {
   ) {
     const existingWeight = assistantContentWeight(existing.content);
     const incomingWeight = assistantContentWeight(incoming.content);
-    if (existingWeight > incomingWeight) return existingEntry;
-    if (incomingWeight > existingWeight) return incomingEntry;
-    return isTerminalAssistantStatus(existing?.status) &&
-      !isTerminalAssistantStatus(incoming?.status)
-      ? existingEntry
-      : incomingEntry;
+    const chosen =
+      existingWeight > incomingWeight
+        ? existingEntry
+        : incomingWeight > existingWeight
+          ? incomingEntry
+          : isTerminalAssistantStatus(existing?.status) &&
+              !isTerminalAssistantStatus(incoming?.status)
+            ? existingEntry
+            : incomingEntry;
+    return preserveAssistantRunDuration(
+      chosen,
+      chosen === existingEntry ? incomingEntry : existingEntry,
+    );
   }
   if (
     existing?.role === "assistant" &&
@@ -492,9 +544,9 @@ function chooseMergedMessageEntry(existingEntry: any, incomingEntry: any): any {
     isTerminalAssistantStatus(existing?.status) &&
     !isTerminalAssistantStatus(incoming?.status)
   ) {
-    return existingEntry;
+    return preserveAssistantRunDuration(existingEntry, incomingEntry);
   }
-  return incomingEntry;
+  return preserveAssistantRunDuration(incomingEntry, existingEntry);
 }
 
 function normalizeMessageEntry(
@@ -1397,6 +1449,17 @@ function foldedRunIdsOf(message: any): string[] {
     : [];
 }
 
+function assistantRunDurationMs(
+  custom: Record<string, unknown>,
+): number | null {
+  const durationMs = custom[ASSISTANT_RUN_DURATION_METADATA_KEY];
+  return typeof durationMs === "number" &&
+    Number.isFinite(durationMs) &&
+    durationMs >= 0
+    ? durationMs
+    : null;
+}
+
 /** Rough size of an assistant message's content, used only to pick the larger
  *  of two representations of the same chunk so a fold can never shrink. */
 function assistantContentWeight(content: unknown): number {
@@ -1519,6 +1582,20 @@ export function foldAssistantTurn(
     turnId,
     foldedRunIds: mergedFolded,
   };
+  const existingDurationMs = assistantRunDurationMs(existingCustom);
+  const incomingDurationMs = assistantRunDurationMs(incomingCustom);
+  const mergedDurationMs = runAlreadyFolded
+    ? existingDurationMs == null
+      ? incomingDurationMs
+      : incomingDurationMs == null
+        ? existingDurationMs
+        : Math.max(existingDurationMs, incomingDurationMs)
+    : existingDurationMs == null && incomingDurationMs == null
+      ? null
+      : (existingDurationMs ?? 0) + (incomingDurationMs ?? 0);
+  if (mergedDurationMs != null) {
+    mergedCustom[ASSISTANT_RUN_DURATION_METADATA_KEY] = mergedDurationMs;
+  }
   // Only the freshest chunk decides whether the turn is still continuing.
   if (incomingCustom.continued !== true) delete mergedCustom.continued;
 

package/corpus/core/src/client/AssistantChat.tsx

@@ -110,6 +110,7 @@ import {
   shouldImportServerThreadData,
   dedupeRepoMessagesById,
   dropEmptyAssistantMessages,
+  withLastAssistantRunDuration,
 } from "./chat/repo-helpers.js";
 import {
   BuilderSetupContent,
@@ -2842,6 +2843,26 @@ const AssistantChatInner = forwardRef<
       ensureMessageMetadata(normalizeThreadRepository(threadRuntime.export())),
     [threadRuntime],
   );
+  const exportPersistableThreadRepo = useCallback(
+    () =>
+      withLastAssistantRunDuration(
+        exportCleanThreadRepo(),
+        showRunningInUI ? null : lastChatRunDurationMs,
+      ),
+    [exportCleanThreadRepo, lastChatRunDurationMs, showRunningInUI],
+  );
+  useEffect(() => {
+    if (showRunningInUI || lastChatRunDurationMs == null) return;
+    const repo = exportCleanThreadRepo();
+    const persisted = withLastAssistantRunDuration(repo, lastChatRunDurationMs);
+    if (persisted === repo) return;
+    threadRuntime.import(ensureMessageMetadata(persisted));
+  }, [
+    exportCleanThreadRepo,
+    lastChatRunDurationMs,
+    showRunningInUI,
+    threadRuntime,
+  ]);
 
   const appendRealtimeVoiceTranscript = useCallback(
     (
@@ -2896,7 +2917,7 @@ const AssistantChatInner = forwardRef<
 
   const cacheCurrentThreadSnapshot = useCallback(() => {
     if (!threadId || messages.length === 0) return;
-    const repo = exportCleanThreadRepo();
+    const repo = exportPersistableThreadRepo();
     const threadData = JSON.stringify(stripBase64FromRepo(repo));
     const { title, preview } = extractThreadMeta(repo);
     writeCachedThreadSnapshot(apiUrl, threadId, {
@@ -2905,7 +2926,7 @@ const AssistantChatInner = forwardRef<
       preview,
       messageCount: messages.length,
     });
-  }, [apiUrl, exportCleanThreadRepo, messages.length, threadId]);
+  }, [apiUrl, exportPersistableThreadRepo, messages.length, threadId]);
 
   useBrowserLayoutEffect(() => {
     if (hasImportedInitialCachedSnapshotRef.current) return;
@@ -3669,7 +3690,7 @@ const AssistantChatInner = forwardRef<
     const timeSinceLastSave = now - lastSaveTimeRef.current;
     if (timeSinceLastSave < 5000) return;
 
-    const repo = exportCleanThreadRepo();
+    const repo = exportPersistableThreadRepo();
     const { title, preview } = extractThreadMeta(repo);
     const threadData = JSON.stringify(stripBase64FromRepo(repo));
     const snapshot = {
@@ -3683,7 +3704,7 @@ const AssistantChatInner = forwardRef<
     savedTitleRef.current = title;
     writeCachedThreadSnapshot(apiUrl, threadId, snapshot);
     onSaveThreadRef.current(threadId, snapshot);
-  }, [apiUrl, exportCleanThreadRepo, messages, isRunning, threadId]);
+  }, [apiUrl, exportPersistableThreadRepo, messages, isRunning, threadId]);
 
   // Persist full thread data after each completed response
   useEffect(() => {
@@ -3691,7 +3712,7 @@ const AssistantChatInner = forwardRef<
     if (isRunning) return;
     if (messages.length === 0) return;
 
-    const repo = exportCleanThreadRepo();
+    const repo = exportPersistableThreadRepo();
 
     if (threadId && onSaveThreadRef.current) {
       // Save to server via the hook callback
@@ -3713,7 +3734,14 @@ const AssistantChatInner = forwardRef<
         sessionStorage.setItem(storageKey, JSON.stringify(repo));
       } catch {}
     }
-  }, [apiUrl, exportCleanThreadRepo, messages, isRunning, threadId, tabId]);
+  }, [
+    apiUrl,
+    exportPersistableThreadRepo,
+    messages,
+    isRunning,
+    threadId,
+    tabId,
+  ]);
 
   useEffect(() => {
     onMessageCountChange?.(messages.length);
@@ -4368,8 +4396,8 @@ const AssistantChatInner = forwardRef<
         }
         reconnectTailOnlyRef.current = false;
       }
-      threadRuntime.cancelRun();
       settleVisibleInterruptedTools();
+      threadRuntime.cancelRun();
       if (typeof window !== "undefined") {
         window.dispatchEvent(
           new CustomEvent("agentNative.chatRunning", {
@@ -4810,7 +4838,7 @@ const AssistantChatInner = forwardRef<
       },
       exportThreadSnapshot() {
         if (messages.length === 0) return null;
-        const repo = exportCleanThreadRepo();
+        const repo = exportPersistableThreadRepo();
         const { title, preview } = extractThreadMeta(repo);
         return {
           threadData: JSON.stringify(repo),
@@ -4822,7 +4850,7 @@ const AssistantChatInner = forwardRef<
     }),
     [
       addToQueue,
-      exportCleanThreadRepo,
+      exportPersistableThreadRepo,
       isRunning,
       messages.length,
       stageComposerContextItem,

package/corpus/core/src/client/chat/message-components.tsx

@@ -73,6 +73,7 @@ import {
   MarkdownText,
   renderMarkdownToClipboardHtml,
 } from "./markdown-renderer.js";
+import { getAssistantRunDurationMs } from "./repo-helpers.js";
 import {
   ToolCallFallback,
   ToolActivityPresentation,
@@ -989,17 +990,22 @@ export function shouldShowAssistantMessageFooter({
 }
 
 export function shouldShowMissingFinalResponse({
+  isCurrentTurnRunning,
   statusIsTerminal,
   hasAssistantText,
   hasUnresolvedTool,
   hasCompletedCustomUi,
 }: {
+  isCurrentTurnRunning: boolean;
   statusIsTerminal: boolean;
   hasAssistantText: boolean;
   hasUnresolvedTool: boolean;
   hasCompletedCustomUi?: boolean;
 }): boolean {
+  // A completed tool can make the latest message look terminal before the
+  // active turn attaches its follow-up text.
   return (
+    !isCurrentTurnRunning &&
     statusIsTerminal &&
     !hasAssistantText &&
     !hasUnresolvedTool &&
@@ -1202,6 +1208,7 @@ export function AssistantMessage() {
   const chatRunning = React.useContext(ChatRunningContext);
   const lastRunDurationMs = React.useContext(ChatRunDurationContext);
   const msg = messageRuntime.getState();
+  const persistedDurationMs = getAssistantRunDurationMs(msg);
   const timestamp = formatMessageTimestamp(msg.createdAt);
   const isLast =
     thread.messages.length > 0 &&
@@ -1215,6 +1222,7 @@ export function AssistantMessage() {
   );
   const hasCustomUi = assistantMessageHasCustomUi(msg.content);
   const showMissingFinalResponse = shouldShowMissingFinalResponse({
+    isCurrentTurnRunning: isLast && chatRunning,
     statusIsTerminal,
     hasAssistantText: responseConnectionText.trim().length > 0,
     hasUnresolvedTool,
@@ -1370,7 +1378,7 @@ export function AssistantMessage() {
                 if (!showSummary) return <>{children}</>;
                 return (
                   <WorkedForSummary
-                    durationMs={capturedDurationMs}
+                    durationMs={capturedDurationMs ?? persistedDurationMs}
                     defaultOpen={hasCustomUi}
                     autoCollapse={animateCollapse && !hasCustomUi}
                   >

package/corpus/core/src/client/chat/repo-helpers.ts

@@ -5,6 +5,8 @@
 // interface covering the shapes both `threadRuntime.export()` and
 // `normalizeThreadRepository()` produce, and use it everywhere instead of `any`.
 
+import { ASSISTANT_RUN_DURATION_METADATA_KEY } from "../../agent/thread-data-builder.js";
+
 export interface RepoMessageStatus {
   type?: string;
   reason?: string;
@@ -56,6 +58,81 @@ export function getRepoMessage(entry: RepoEntry): RepoMessage | null {
   return (entry?.message ?? entry) as RepoMessage | null;
 }
 
+export function getAssistantRunDurationMs(
+  message:
+    | {
+        metadata?: unknown;
+      }
+    | null
+    | undefined,
+): number | null {
+  const metadata =
+    message?.metadata && typeof message.metadata === "object"
+      ? (message.metadata as Record<string, unknown>)
+      : null;
+  const custom =
+    metadata?.custom && typeof metadata.custom === "object"
+      ? (metadata.custom as Record<string, unknown>)
+      : null;
+  const durationMs = custom?.[ASSISTANT_RUN_DURATION_METADATA_KEY];
+  return typeof durationMs === "number" &&
+    Number.isFinite(durationMs) &&
+    durationMs >= 0
+    ? durationMs
+    : null;
+}
+
+export function withLastAssistantRunDuration<T extends NormalizedRepo>(
+  repo: T,
+  durationMs: number | null | undefined,
+): T {
+  if (
+    !Array.isArray(repo.messages) ||
+    typeof durationMs !== "number" ||
+    !Number.isFinite(durationMs) ||
+    durationMs < 0
+  ) {
+    return repo;
+  }
+
+  let messageIndex = -1;
+  for (let index = repo.messages.length - 1; index >= 0; index -= 1) {
+    if (getRepoMessage(repo.messages[index]!)?.role === "assistant") {
+      messageIndex = index;
+      break;
+    }
+  }
+  if (messageIndex < 0) return repo;
+
+  const entry = repo.messages[messageIndex]!;
+  const message = getRepoMessage(entry);
+  if (!message || getAssistantRunDurationMs(message) === durationMs) {
+    return repo;
+  }
+
+  const metadata = message.metadata ?? {};
+  const custom =
+    metadata.custom && typeof metadata.custom === "object"
+      ? (metadata.custom as Record<string, unknown>)
+      : {};
+  const nextMessage: RepoMessage = {
+    ...message,
+    metadata: {
+      ...metadata,
+      custom: {
+        ...custom,
+        [ASSISTANT_RUN_DURATION_METADATA_KEY]: durationMs,
+      },
+    },
+  };
+  const messages = repo.messages.slice();
+  messages[messageIndex] =
+    entry.message === undefined
+      ? (nextMessage as RepoEntry)
+      : { ...entry, message: nextMessage };
+  return { ...repo, messages };
+}
+
 /**
  * Collapse duplicate message ids before a repository is handed to
  * `threadRuntime.import()`. assistant-ui's `MessageRepository` throws

package/corpus/core/src/client/chat/tool-call-display.tsx

@@ -25,7 +25,10 @@ import React, {
   useRef,
 } from "react";
 
-import type { A2AAgentActivitySnapshot } from "../../a2a/activity.js";
+import type {
+  A2AAgentActivitySnapshot,
+  A2AAgentActivityToolCall,
+} from "../../a2a/activity.js";
 import type { ActionChatUIConfig } from "../../action-ui.js";
 import type { AgentMcpAppPayload } from "../../mcp-client/app-result.js";
 import { AgentTaskCard } from "../AgentTaskCard.js";
@@ -100,11 +103,13 @@ export function ToolActivityPresentation({
   toolName,
   isRunning,
   isActiveTail,
+  suppressLongRunningHint = false,
   children,
 }: {
   toolName: string;
   isRunning: boolean;
   isActiveTail: boolean;
+  suppressLongRunningHint?: boolean;
   children: React.ReactNode;
 }) {
   const [showLongRunningHint, setShowLongRunningHint] = useState(false);
@@ -114,7 +119,7 @@ export function ToolActivityPresentation({
   const [animateEntry] = useState(isActiveTail);
 
   useEffect(() => {
-    if (!isRunning) {
+    if (!isRunning || suppressLongRunningHint) {
       setShowLongRunningHint(false);
       return;
     }
@@ -123,7 +128,7 @@ export function ToolActivityPresentation({
       setShowLongRunningHint(true);
     }, TOOL_LONG_RUNNING_HINT_DELAY_MS);
     return () => window.clearTimeout(timeout);
-  }, [isRunning, toolName]);
+  }, [isRunning, suppressLongRunningHint, toolName]);
 
   return (
     <div
@@ -584,6 +589,9 @@ export function ToolCallDisplay({
       toolName={toolName}
       isRunning={isRunning}
       isActiveTail={showActiveTail}
+      suppressLongRunningHint={
+        toolName === "call-agent" || toolName.startsWith("agent:")
+      }
     >
       {children}
     </ToolActivityPresentation>
@@ -955,18 +963,8 @@ function AgentCallCell({
               />
             )}
             {tool && (
-              <ToolCallDisplay
-                toolName={tool.name}
-                args={{}}
-                argsText=""
-                isRunning={tool.status === "running"}
-                result={
-                  tool.status === "failed"
-                    ? "Failed"
-                    : tool.status === "completed"
-                      ? "Done"
-                      : undefined
-                }
+              <AgentActivityToolCallRow
+                tool={tool}
                 isActiveTail={
                   isRunning && index === activity.toolCalls.length - 1
                 }
@@ -1051,6 +1049,44 @@ function AgentCallCell({
   );
 }
 
+function AgentActivityToolCallRow({
+  tool,
+  isActiveTail,
+}: {
+  tool: A2AAgentActivityToolCall;
+  isActiveTail: boolean;
+}) {
+  const isRunning = tool.status === "running";
+  const ToolIcon = resolveToolIcon(tool.name);
+
+  return (
+    <ToolActivityPresentation
+      toolName={tool.name}
+      isRunning={isRunning}
+      isActiveTail={isActiveTail}
+      suppressLongRunningHint
+    >
+      <div className="my-0.5 flex w-full items-center gap-1.5 rounded-md py-0.5 text-left text-[13px] text-muted-foreground">
+        <span className="flex size-4 shrink-0 items-center justify-center">
+          {isRunning ? (
+            <IconLoader2 className="size-3.5 animate-spin" />
+          ) : (
+            <ToolIcon className="size-3.5" />
+          )}
+        </span>
+        <span
+          className={cn(
+            "min-w-0 truncate font-normal",
+            isActiveTail && "agent-running-shimmer",
+          )}
+        >
+          {humanizeToolName(tool.name)}
+        </span>
+      </div>
+    </ToolActivityPresentation>
+  );
+}
+
 // ─── ToolCallFallback ──────────────────────────────────────────────────────────
 
 export function ToolCallFallback({

package/corpus/core/src/scripts/agent-engines/list-agent-engines.ts

@@ -13,6 +13,7 @@ import {
   isAgentEnginePackageInstalled,
   isStoredEngineUsableForRequest,
   normalizeModelForEngine,
+  resolveEnginePreservesCustomModels,
 } from "../../agent/engine/index.js";
 import type { ActionTool } from "../../agent/types.js";
 import { getSetting } from "../../settings/index.js";
@@ -82,11 +83,18 @@ export async function run(args: Record<string, string> = {}): Promise<string> {
         ? current?.model
         : undefined;
   const currentEngineName = currentEntry?.name ?? "anthropic";
+  // Resolve the OpenAI-compatible-endpoint capability so a custom gateway model
+  // is reported as-is instead of being normalized to the engine default — the
+  // read-side counterpart of the same fix in set-/manage-agent-engine.
+  const preserveCustomModels = currentEntry
+    ? await resolveEnginePreservesCustomModels(currentEntry)
+    : false;
   const currentModel =
     currentEntry && !envUnavailable
       ? normalizeModelForEngine(
           currentEntry,
           currentModelCandidate ?? currentEntry.defaultModel,
+          { preserveCustomModels },
         )
       : (currentModelCandidate ?? DEFAULT_MODEL);
   const result = {

package/corpus/core/src/scripts/agent-engines/manage-agent-engine.ts

@@ -16,6 +16,7 @@ import {
   getAgentEngineEntry,
   isAgentEnginePackageInstalled,
   normalizeModelForEngine,
+  resolveEnginePreservesCustomModels,
   registerBuiltinEngines,
 } from "../../agent/engine/index.js";
 import type { ActionTool } from "../../agent/types.js";
@@ -113,7 +114,10 @@ async function runSetAppDefault(args: Record<string, string>): Promise<string> {
   if (!isAgentEnginePackageInstalled(entry)) {
     return `Error: Engine "${engine}" requires optional packages that are not installed in this app. Run: pnpm add ${entry.installPackage}`;
   }
-  const normalizedModel = normalizeModelForEngine(entry, model);
+  const preserveCustomModels = await resolveEnginePreservesCustomModels(entry);
+  const normalizedModel = normalizeModelForEngine(entry, model, {
+    preserveCustomModels,
+  });
 
   const ctx = currentContext();
   const canUpdate = await canUpdateAgentAppModelDefaultSettings(

package/corpus/core/src/scripts/agent-engines/set-agent-engine.ts

@@ -8,6 +8,7 @@ import {
   isAgentEnginePackageInstalled,
   isStoredEngineUsableForRequest,
   normalizeModelForEngine,
+  resolveEnginePreservesCustomModels,
   registerBuiltinEngines,
 } from "../../agent/engine/index.js";
 import type { ActionTool } from "../../agent/types.js";
@@ -54,7 +55,14 @@ export async function run(args: Record<string, string>): Promise<string> {
   }
 
   const requestedModel = model ?? entry.defaultModel;
-  const resolvedModel = normalizeModelForEngine(entry, requestedModel);
+  // A static registry entry can't carry the runtime `preserveCustomModels`
+  // flag, so resolve the OpenAI-compatible-endpoint capability here and pass it
+  // through — otherwise a gateway model (e.g. an Ollama id) is rewritten to the
+  // engine default on save.
+  const preserveCustomModels = await resolveEnginePreservesCustomModels(entry);
+  const resolvedModel = normalizeModelForEngine(entry, requestedModel, {
+    preserveCustomModels,
+  });
 
   const usable = await isStoredEngineUsableForRequest(
     { engine: engineName },

package/corpus/core/src/server/agent-chat/framework-prompts.ts

@@ -413,7 +413,9 @@ export function generateCorpusToolsPrompt(
 
 Available corpus-capable tools: ${available.join(", ")}.
 
-For broad provider searches, raw API access, multi-page cohorts, cross-source joins, classification/counting over records, or absence-sensitive answers, do not stop at a bounded shortcut action. Use the provider's broad API/search/list surface, fetch every relevant page or an explicit bounded cohort, stage/save large responses when needed, and reduce the corpus with durable corpus jobs, staged-dataset queries, or code execution.
+This workflow does not apply to ordinary structured lookups, bounded aggregates, or counts grouped over one known source. For those requests, use the single most directly authoritative source, run one bounded query, and answer as soon as it succeeds. Do not cross-check or expand into a corpus workflow unless the user asks for multiple sources, exhaustive unstructured-record coverage, or an absence claim.
+
+For broad provider searches, raw API access, multi-page cohorts, cross-source joins, classification over unstructured records, or absence-sensitive answers, do not stop at a bounded shortcut action. Use the provider's broad API/search/list surface, fetch every relevant page or an explicit bounded cohort, stage/save large responses when needed, and reduce the corpus with durable corpus jobs, staged-dataset queries, or code execution.
 
 When \`provider-corpus-job\` is available, prefer it for transcript/message/ticket/issue/document scans that may exceed one turn, need provider-side backoff, or need a defensible "not found" conclusion. Use operation="start" with mode="paginated-search" for any paginated provider endpoint, or mode="batch-search" when a prior cohort of ids/records must feed a second provider endpoint. Continue paused jobs with operation="continue" until status is completed or quota_wait, then read operation="results". In run-code, prefer providerFetchAll() for short cursor/page/offset pagination and providerRequest() when response status, headers, or truncation metadata matters. Report source, filters, row counts, pagination/truncation, failed pages, quota_wait times, and remaining gaps.`;
 }

package/corpus/core/src/server/agent-chat/plugin-options.ts

@@ -216,6 +216,15 @@ export interface AgentChatPluginOptions {
    * tool-search unless the app explicitly includes them in this list.
    */
   initialToolNames?: string[];
+  /**
+   * Controls whether broad provider/corpus tools and their workflow prompt are
+   * loaded into the first model request. Use `"lazy"` for apps that can answer
+   * ordinary lookups from a compact curated tool set; corpus tools remain
+   * discoverable through tool-search and the full retry surface.
+   *
+   * @default "initial"
+   */
+  corpusTools?: "initial" | "lazy";
   /**
    * Use a compact system prompt with on-demand context loading. The system
    * prompt includes essential behavioral rules and action signatures, but

package/corpus/core/src/server/agent-chat-plugin.ts

@@ -1179,8 +1179,10 @@ export function createAgentChatPlugin(
       // that never emit the corpus prompt (no provider/run-code tools
       // registered), so this never silently expands the initial set for
       // apps that don't teach these tools by name.
-      const corpusToolNames =
-        corpusToolNamesTaughtByPrompt(corpusPromptRegistry);
+      const loadCorpusToolsInitially = options?.corpusTools !== "lazy";
+      const corpusToolNames = loadCorpusToolsInitially
+        ? corpusToolNamesTaughtByPrompt(corpusPromptRegistry)
+        : [];
       const effectiveInitialToolNames =
         corpusToolNames.length > 0
           ? [...new Set([...templateInitialToolNames, ...corpusToolNames])]
@@ -1907,7 +1909,9 @@ export function createAgentChatPlugin(
       // Dev: actions are invoked via bash — emit `pnpm action name --arg <type>`
       //      and include discoveredActions too, since those are also missing
       //      from the dev tool registry.
-      const corpusToolsPrompt = generateCorpusToolsPrompt(corpusPromptRegistry);
+      const corpusToolsPrompt = loadCorpusToolsInitially
+        ? generateCorpusToolsPrompt(corpusPromptRegistry)
+        : "";
       const prodActionsPrompt =
         generateActionsPrompt(
           templateScripts,
@@ -2250,6 +2254,11 @@ export function createAgentChatPlugin(
                   typeof run.turnId === "string" && run.turnId
                     ? run.turnId
                     : undefined,
+                runDurationMs:
+                  typeof run.startedAt === "number" &&
+                  Number.isFinite(run.startedAt)
+                    ? Math.max(0, Date.now() - run.startedAt)
+                    : undefined,
               },
             );
             if (!assistantMsg) {