Agent publishing to the whole organization, trust levels, usage statistics (v2.4.0) (#153)

* feat(agents): organization-wide publishing, trust levels and usage statistics

Any user can now publish their agents to the whole organization via a
switch in the agent's "Access to Agent" section (publishing supersedes
Entra-group sharing, which is deactivated while published).

- Trust classification: Verified / Community badges on agent cards.
  Legacy admin-published agents count as Verified (lazily materialized
  on save). Classification actions (verify/downgrade/unpublish) are
  restricted to members of the Entra ID group in AGENT_VERIFIER_GROUP_ID
  (Graph checkMemberGroups, fail-closed, re-checked server-side).
- Per-agent usage statistics (chats, interactions, tokens) as atomic
  Cosmos counters: AGENT_STATS docs under a sentinel partition, updated
  via Patch API increments from CreatePersonaChat and persistThread
  (threads now carry personaId). No time series; single-partition reads.
- Home page shows favorites only, with "Browse all agents"; the Agents
  page gains sorting (most used default / recently updated / newest /
  name) and a trust filter.
- Fix: UpsertPersona no longer overwrites createdAt on every edit
  (root cause of "last edited agent always on top"); adds updatedAt.

* feat(models): GPT-5.6 Sol/Terra/Luna, rolling 7-day budget window; release v2.4.0

- Add GPT-5.6 family (Sol flagship / Terra balanced / Luna fast); Sol
  becomes the default model, Luna replaces GPT-5.4 Mini as the cost-cap
  downgrade target. GPT-5.3 Chat removed.
- Budget window semantics: GetWeeklyUsage now computes an exact rolling
  7-calendar-day window (-6 with inclusive compare); usage menu shows
  "Last 7 Days (rolling)" instead of a fixed Friday reset; downgrade
  message wording matches.
- Release v2.4.0: changelog entries dated, package version bumped.

Author: saoc90

src/.env.example

@@ -14,7 +14,9 @@ AZURE_OPENAI_API_EMBEDDINGS_DEPLOYMENT_NAME=text-embedding-3-large
 
 # New v1 API model deployments for Responses API
 AZURE_OPENAI_API_GPT52_DEPLOYMENT_NAME=gpt-5.2
-AZURE_OPENAI_API_GPT53_CHAT_DEPLOYMENT_NAME=gpt-5.3-chat
+AZURE_OPENAI_API_GPT56_SOL_DEPLOYMENT_NAME=gpt-5.6-sol
+AZURE_OPENAI_API_GPT56_TERRA_DEPLOYMENT_NAME=gpt-5.6-terra
+AZURE_OPENAI_API_GPT56_LUNA_DEPLOYMENT_NAME=gpt-5.6-luna
 AZURE_OPENAI_API_GPT55_DEPLOYMENT_NAME=gpt-5.5
 AZURE_OPENAI_API_GPT54_DEPLOYMENT_NAME=gpt-5.4
 AZURE_OPENAI_API_GPT54_MINI_DEPLOYMENT_NAME=gpt-5.4-mini
@@ -27,7 +29,7 @@ AZURE_OPENAI_GPT_IMAGE_DEPLOYMENT_NAME=gpt-image-1
 # Foundry-hosted low-cost models (DeepSeek/Kimi) used as automatic downgrade
 # targets. OpenAI-compatible Chat Completions on the Foundry host's /openai/v1
 # surface; Bearer API-key auth. Leave unset to disable Foundry downgrade
-# targets (gpt-5.4-mini still works as a hardCapEligible Azure fallback).
+# targets (gpt-5.6-luna still works as a hardCapEligible Azure fallback).
 FOUNDRY_OPENAI_BASE_URL=https://<resource>.services.ai.azure.com/openai/v1
 FOUNDRY_API_KEY=
 FOUNDRY_DEEPSEEK_DEPLOYMENT_NAME=DeepSeek-V4-Pro
@@ -66,6 +68,11 @@ AZURE_OPENAI_VISION_API_VERSION=2023-12-01-preview
 # Update your admin email addresses - comma separated
 ADMIN_EMAIL_ADDRESS=you@email.com,you2@email.com
 
+# Agent trust management: object ID of the Entra ID group whose members may
+# verify/downgrade/unpublish organization-published agents. May be a
+# security group. Leave empty to disable trust actions for everyone.
+AGENT_VERIFIER_GROUP_ID=
+
 # Identity provider is optional if you are running in development mode locally (npm run dev)
 AUTH_GITHUB_ID=
 AUTH_GITHUB_SECRET=

src/__tests__/INVENTORY.md

@@ -318,7 +318,7 @@ This is the largest feature folder with complex state management and streaming.
 
 **Models:**
 - `chat-services/models.ts`
-  - `ChatModel` type (gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5.3-chat)
+  - `ChatModel` type (gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5, gpt-5.4, gpt-5.4-mini)
   - `MODEL_CONFIGS` record with pricing, context window, reasoning support
   - `ChatThreadModel`, `ChatMessageModel`, `ChatDocumentModel`, `AttachedFileModel`
   - `UserPrompt`, `AzureChatCompletion` response types
@@ -911,7 +911,7 @@ This folder contains shadcn/ui and custom UI primitives. Most are pure presentat
 
 **Completions API** (main LLM)
 - Used in: chat-api-response.ts, openai-responses-stream.ts
-- Models: gpt-5.5, gpt-5.4, gpt-5.4-mini, gpt-5.3-chat
+- Models: gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5, gpt-5.4, gpt-5.4-mini
 - Test strategy: Mock streaming response
 
 **Embeddings API**

src/__tests__/e2e-fakes/cosmos.ts

@@ -60,6 +60,10 @@ class InMemoryContainer {
             this.docs[idx][key] = op.value;
           } else if (op.op === "remove") {
             delete this.docs[idx][key];
+          } else if (op.op === "incr") {
+            const current = this.docs[idx][key];
+            this.docs[idx][key] =
+              (typeof current === "number" ? current : 0) + (op.value ?? 0);
           }
         }
         return { resource: this.docs[idx] };

src/app/(authenticated)/api/chat/route.ts

@@ -472,6 +472,7 @@ export async function POST(req: Request) {
           modelConfig,
           fallbackInfo: fallbackInfo.fellBack ? fallbackInfo : undefined,
           reasoningDurationMs: computeReasoningDurationMs(),
+          personaId: ctx.thread.isTemporary ? undefined : ctx.thread.personaId,
         });
       } catch (err) {
         logError("/api/chat onAbort persist failed", {
@@ -521,6 +522,7 @@ export async function POST(req: Request) {
           fallbackInfo: fallbackInfo.fellBack ? fallbackInfo : undefined,
           streamError: lastStreamError,
           reasoningDurationMs: computeReasoningDurationMs(),
+          personaId: ctx.thread.isTemporary ? undefined : ctx.thread.personaId,
         });
         // Generate a thread title from the first user message. ctx.history
         // has just the user message we appended in loadThreadContext when
@@ -578,6 +580,9 @@ export async function POST(req: Request) {
             fallbackInfo: fallbackInfo.fellBack ? fallbackInfo : undefined,
             streamError: lastStreamError,
             reasoningDurationMs: computeReasoningDurationMs(),
+            personaId: ctx.thread.isTemporary
+              ? undefined
+              : ctx.thread.personaId,
           });
         } catch (retryErr) {
           logError("/api/chat persistAssistantFromFinishEvent retry failed", {

src/app/(authenticated)/chat/page.tsx

@@ -1,18 +1,20 @@
 import { userHashedId } from "@/features/auth-page/helpers";
 import { ChatHome } from "@/features/chat-home-page/chat-home";
+import { GetAllAgentStats } from "@/features/common/services/agent-stats-service";
 import { FindAllNewsArticles } from "@/features/common/services/news-service/news-service";
 import { FindAllExtensionForCurrentUser } from "@/features/extensions-page/extension-services/extension-service";
 import { GetUserFavoriteAgents } from "@/features/persona-page/persona-services/agent-favorite-service";
 import { FindAllPersonaForCurrentUser } from "@/features/persona-page/persona-services/persona-service";
 import { DisplayError } from "@/features/ui/error/display-error";
 
 export default async function Home() {
-  const [personaResponse, extensionResponse, newsResponse, favoriteAgentIds, currentUserId] = await Promise.all([
+  const [personaResponse, extensionResponse, newsResponse, favoriteAgentIds, currentUserId, agentStats] = await Promise.all([
     FindAllPersonaForCurrentUser(),
     FindAllExtensionForCurrentUser(),
     FindAllNewsArticles(),
     GetUserFavoriteAgents(),
     userHashedId(),
+    GetAllAgentStats(),
   ]);
 
   if (personaResponse.status !== "OK") {
@@ -34,6 +36,7 @@ export default async function Home() {
       news={newsResponse.response}
       favoriteAgentIds={favoriteAgentIds}
       currentUserId={currentUserId}
+      agentStats={agentStats}
     />
   );
 }

src/app/(authenticated)/persona/page.tsx

@@ -1,18 +1,28 @@
 import { userHashedId } from "@/features/auth-page/helpers";
+import { GetAllAgentStats } from "@/features/common/services/agent-stats-service";
 import { FindAllExtensionForCurrentUser } from "@/features/extensions-page/extension-services/extension-service";
 import { ChatPersonaPage } from "@/features/persona-page/persona-page";
 import { GetUserFavoriteAgents } from "@/features/persona-page/persona-services/agent-favorite-service";
+import { IsAgentVerifier } from "@/features/persona-page/persona-services/agent-trust-service";
 import { FindAllPersonaForCurrentUser } from "@/features/persona-page/persona-services/persona-service";
 import { DisplayError } from "@/features/ui/error/display-error";
 
 export default async function Home() {
-  const [personasResponse, extensionsResponse, favoriteIds, currentUserId] =
-    await Promise.all([
-      FindAllPersonaForCurrentUser(),
-      FindAllExtensionForCurrentUser(),
-      GetUserFavoriteAgents(),
-      userHashedId(),
-    ]);
+  const [
+    personasResponse,
+    extensionsResponse,
+    favoriteIds,
+    currentUserId,
+    agentStats,
+    isVerifier,
+  ] = await Promise.all([
+    FindAllPersonaForCurrentUser(),
+    FindAllExtensionForCurrentUser(),
+    GetUserFavoriteAgents(),
+    userHashedId(),
+    GetAllAgentStats(),
+    IsAgentVerifier(),
+  ]);
   if (personasResponse.status !== "OK") {
     return <DisplayError errors={personasResponse.errors} />;
   }
@@ -25,6 +35,8 @@ export default async function Home() {
       extensions={extensionsResponse.response}
       initialFavoriteIds={favoriteIds}
       currentUserId={currentUserId}
+      agentStats={agentStats}
+      isVerifier={isVerifier}
     />
   );
 }

src/features/chat-home-page/chat-home.tsx

@@ -1,5 +1,6 @@
 "use client";
 
+import { AgentStatsModel } from "@/features/common/services/agent-stats-models";
 import { ExtensionModel } from "@/features/extensions-page/extension-services/models";
 import { AgentList } from "@/features/persona-page/agent-list";
 import { AddNewPersona } from "@/features/persona-page/add-new-persona";
@@ -21,6 +22,7 @@ interface ChatPersonaProps {
   news: NewsArticleModel[];
   favoriteAgentIds: string[];
   currentUserId: string;
+  agentStats?: Record<string, AgentStatsModel>;
 }
 
 const FeedbackButton = ({ feedBackLink }: { feedBackLink: string }) => (
@@ -100,7 +102,7 @@ const ArticlesSection = ({
   </div>
 );
 
-export const ChatHome: FC<ChatPersonaProps> = ({ personas, extensions, news, favoriteAgentIds, currentUserId }) => {
+export const ChatHome: FC<ChatPersonaProps> = ({ personas, extensions, news, favoriteAgentIds, currentUserId, agentStats }) => {
   const [showChangelog, setShowChangelog] = useState<boolean>(false);
 
   return (
@@ -134,6 +136,8 @@ export const ChatHome: FC<ChatPersonaProps> = ({ personas, extensions, news, fav
                 personas={personas}
                 initialFavoriteIds={favoriteAgentIds}
                 currentUserId={currentUserId}
+                agentStats={agentStats}
+                favoritesOnly
               />
             </>
           )}

src/features/chat-page/__tests__/chat-store-factory.test.ts

@@ -68,12 +68,12 @@ describe("createChatStore — isolation", () => {
     storeA.subscribe(listenerA);
     storeB.subscribe(listenerB);
 
-    storeA.getState().setSelectedModel("gpt-5.3-chat");
+    storeA.getState().setSelectedModel("gpt-5.4");
 
     expect(listenerA).toHaveBeenCalledOnce();
     expect(listenerB).not.toHaveBeenCalled();
-    expect(storeA.getState().selectedModel).toBe("gpt-5.3-chat");
-    expect(storeB.getState().selectedModel).not.toBe("gpt-5.3-chat");
+    expect(storeA.getState().selectedModel).toBe("gpt-5.4");
+    expect(storeB.getState().selectedModel).not.toBe("gpt-5.4");
   });
 
   // Tool-call-history restoration was deleted with the dead Zustand subsystem

src/features/chat-page/chat-services/chat-api/model-selection.ts

@@ -149,7 +149,10 @@ export async function resolveModelAndLimits(
           reason: "cap",
           originalModel: selectedModel,
           fallbackModel: target,
-          message: `${budget.window === "weekly" ? "Weekly" : "Daily"} cost budget reached. Using ${targetConfig.name} until it resets.`,
+          message:
+            budget.window === "weekly"
+              ? `Rolling 7-day cost budget reached. Using ${targetConfig.name} until your usage from the last 7 days drops below the cap.`
+              : `Daily cost budget reached. Using ${targetConfig.name} until it resets.`,
           limitType: "cost",
           currentUsage: budget.currentUsd ?? 0,
           limit: budget.limitUsd ?? 0,

src/features/chat-page/chat-services/chat-api/persist-assistant.ts

@@ -28,6 +28,7 @@ import type {
 import { createIdGenerator } from "ai";
 import { userHashedId } from "@/features/auth-page/helpers";
 import { logError, logInfo, logWarn } from "@/features/common/services/logger";
+import { RecordAgentInteraction } from "@/features/common/services/agent-stats-service";
 import { IncrementUsage } from "@/features/common/services/usage-service";
 import {
   reportPromptTokens,
@@ -148,6 +149,8 @@ export interface PersistPayload {
   };
   /** Token usage from the top-level streamText finish callback */
   usage: UsagePayload;
+  /** Agent (persona) the thread was started from — attributes per-agent stats. */
+  personaId?: string;
 }
 
 // ---------------------------------------------------------------------------
@@ -173,6 +176,7 @@ export async function persistThread({
   messages,
   modelConfig,
   usage,
+  personaId,
 }: PersistPayload): Promise<void> {
   const userId = await userHashedId();
 
@@ -336,6 +340,23 @@ export async function persistThread({
     })
   );
 
+  // Per-agent usage counters (atomic Cosmos Patch increments on the
+  // agent-stats doc). Only threads started from an agent carry a personaId.
+  // Note: aborted turns arrive here with zero usage and still count as one
+  // interaction — accepted, the user did see a partial answer.
+  if (personaId) {
+    RecordAgentInteraction(personaId, {
+      inputTokens,
+      outputTokens,
+      cachedTokens,
+    }).catch((err: unknown) =>
+      logError("Failed to record agent interaction", {
+        personaId,
+        error: err instanceof Error ? err.message : String(err),
+      })
+    );
+  }
+
   // Emit App Insights custom metrics (OpenTelemetry) consumed by the
   // "Bühler GPT Usage" monitoring workbook. These were dropped when the
   // legacy orchestrator stack was removed (commit 8f818d6); the streamText
@@ -447,6 +468,8 @@ export interface PersistAssistantFromFinishParams<TOOLS extends ToolSet> {
    * hidden behind boilerplate.
    */
   streamError?: { message: string; name?: string };
+  /** Agent (persona) the thread was started from — attributes per-agent stats. */
+  personaId?: string;
 }
 
 /**
@@ -463,6 +486,7 @@ export async function persistAssistantFromFinishEvent<TOOLS extends ToolSet>({
   messageId,
   streamError,
   reasoningDurationMs,
+  personaId,
 }: PersistAssistantFromFinishParams<TOOLS>): Promise<void> {
   // Detect an empty finish — Azure content-filter trips, aborted streams
   // before any output, or a model error that resolves without text/tools.
@@ -582,5 +606,6 @@ export async function persistAssistantFromFinishEvent<TOOLS extends ToolSet>({
       outputTokens: usageDetails.outputTokens ?? 0,
       cachedTokens,
     },
+    personaId,
   });
 }