feat(33-03): real AI streaming with memory injection + assistant handoff

Replace streamEcho with Puter proxy AI call, inject memory facts as
system message, append memory after each turn. Assistant-to-PM handoff
creates new conversation with context summary.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

server/src/__tests__/33-assistant-handoff.test.ts

@@ -0,0 +1,193 @@
+// [nexus] Tests for assistant handoff route (Plan 33-03)
+import { describe, it, expect, vi, beforeEach } from "vitest";
+
+// ---------------------------------------------------------------------------
+// Mock chatService so we can track calls without a real DB
+// ---------------------------------------------------------------------------
+
+const mockGetConversation = vi.fn();
+const mockListMessages = vi.fn();
+const mockCreateConversation = vi.fn();
+const mockAddSystemMessage = vi.fn();
+
+vi.mock("../services/chat.js", () => ({
+  chatService: () => ({
+    getConversation: mockGetConversation,
+    listMessages: mockListMessages,
+    createConversation: mockCreateConversation,
+    addSystemMessage: mockAddSystemMessage,
+  }),
+}));
+
+// Mock authz — assertBoard is a no-op in tests
+vi.mock("../routes/authz.js", () => ({
+  assertBoard: vi.fn(),
+}));
+
+// ---------------------------------------------------------------------------
+// Import the route factory under test
+// ---------------------------------------------------------------------------
+const { buildHandoffSummary, assistantHandoffRoutes } = await import("../routes/assistant-handoff.js");
+
+// ---------------------------------------------------------------------------
+// Unit tests for buildHandoffSummary (pure function)
+// ---------------------------------------------------------------------------
+
+describe("buildHandoffSummary", () => {
+  it("concatenates user messages into a summary", () => {
+    const messages = [
+      { role: "user", content: "Hello world" },
+      { role: "assistant", content: "Hi there" },
+      { role: "user", content: "Can you build a login page?" },
+    ];
+    const summary = buildHandoffSummary(messages as Array<{ role: string; content: string }>);
+    expect(summary).toContain("Hello world");
+    expect(summary).toContain("Can you build a login page?");
+    expect(summary).not.toContain("Hi there"); // assistant message excluded
+  });
+
+  it("excludes non-user messages (assistant, system)", () => {
+    const messages = [
+      { role: "system", content: "You are a helpful assistant" },
+      { role: "user", content: "Build me a dashboard" },
+      { role: "assistant", content: "Sure!" },
+    ];
+    const summary = buildHandoffSummary(messages as Array<{ role: string; content: string }>);
+    expect(summary).toContain("Build me a dashboard");
+    expect(summary).not.toContain("You are a helpful assistant");
+    expect(summary).not.toContain("Sure!");
+  });
+
+  it("caps summary at 1500 chars", () => {
+    const longContent = "A".repeat(600);
+    const messages = [
+      { role: "user", content: longContent },
+      { role: "user", content: longContent },
+      { role: "user", content: longContent },
+    ];
+    const summary = buildHandoffSummary(messages as Array<{ role: string; content: string }>);
+    expect(summary.length).toBeLessThanOrEqual(1500);
+  });
+
+  it("returns empty string if no user messages", () => {
+    const messages = [
+      { role: "assistant", content: "Hello!" },
+      { role: "system", content: "Context here" },
+    ];
+    const summary = buildHandoffSummary(messages as Array<{ role: string; content: string }>);
+    expect(summary).toBe("");
+  });
+});
+
+// ---------------------------------------------------------------------------
+// Integration-style tests for the route handler
+// ---------------------------------------------------------------------------
+
+describe("POST /conversations/:id/assistant-handoff", () => {
+  beforeEach(() => {
+    vi.clearAllMocks();
+
+    mockGetConversation.mockResolvedValue({
+      id: "conv-123",
+      companyId: "company-abc",
+      agentId: null,
+      title: "Test conversation",
+    });
+
+    mockListMessages.mockResolvedValue({
+      items: [
+        { role: "user", content: "I want to build a todo app" },
+        { role: "assistant", content: "Great idea! Let me help." },
+        { role: "user", content: "It should have drag-and-drop" },
+      ],
+      hasMore: false,
+      nextCursor: null,
+    });
+
+    mockCreateConversation.mockResolvedValue({
+      id: "new-conv-999",
+      companyId: "company-abc",
+      title: "Project from assistant chat",
+      agentId: null,
+    });
+
+    mockAddSystemMessage.mockResolvedValue({
+      id: "msg-handoff-1",
+      role: "system",
+      content: "summary",
+      messageType: "handoff_context",
+    });
+  });
+
+  it("creates a new conversation with handoff_context system message", async () => {
+    const router = assistantHandoffRoutes({} as any);
+
+    // Find the POST handler for /conversations/:id/assistant-handoff
+    const layer = (router as any).stack.find(
+      (l: any) => l.route?.path === "/conversations/:id/assistant-handoff" && l.route?.methods?.post,
+    );
+    expect(layer).toBeDefined();
+
+    const handler = layer.route.stack[0].handle;
+    const req = { params: { id: "conv-123" }, actor: { type: "board" } } as any;
+    const res = { json: vi.fn(), status: vi.fn().mockReturnThis() } as any;
+
+    await handler(req, res);
+
+    expect(mockGetConversation).toHaveBeenCalledWith("conv-123");
+    expect(mockListMessages).toHaveBeenCalledWith("conv-123", { limit: 20 });
+    expect(mockCreateConversation).toHaveBeenCalledWith(
+      "company-abc",
+      expect.objectContaining({ title: "Project from assistant chat" }),
+    );
+    expect(mockAddSystemMessage).toHaveBeenCalledWith(
+      "new-conv-999",
+      expect.objectContaining({ messageType: "handoff_context" }),
+    );
+    expect(res.json).toHaveBeenCalledWith({ targetConversationId: "new-conv-999" });
+  });
+
+  it("builds summary from user messages only", async () => {
+    const router = assistantHandoffRoutes({} as any);
+    const layer = (router as any).stack.find(
+      (l: any) => l.route?.path === "/conversations/:id/assistant-handoff" && l.route?.methods?.post,
+    );
+    const handler = layer.route.stack[0].handle;
+    const req = { params: { id: "conv-123" }, actor: { type: "board" } } as any;
+    const res = { json: vi.fn(), status: vi.fn().mockReturnThis() } as any;
+
+    await handler(req, res);
+
+    const callArgs = mockAddSystemMessage.mock.calls[0];
+    const systemMessageContent = callArgs[1].content;
+    expect(systemMessageContent).toContain("I want to build a todo app");
+    expect(systemMessageContent).toContain("It should have drag-and-drop");
+    expect(systemMessageContent).not.toContain("Great idea! Let me help.");
+  });
+
+  it("caps summary content at 1500 chars in the system message", async () => {
+    const longContent = "X".repeat(600);
+    mockListMessages.mockResolvedValue({
+      items: [
+        { role: "user", content: longContent },
+        { role: "user", content: longContent },
+        { role: "user", content: longContent },
+      ],
+      hasMore: false,
+      nextCursor: null,
+    });
+
+    const router = assistantHandoffRoutes({} as any);
+    const layer = (router as any).stack.find(
+      (l: any) => l.route?.path === "/conversations/:id/assistant-handoff" && l.route?.methods?.post,
+    );
+    const handler = layer.route.stack[0].handle;
+    const req = { params: { id: "conv-123" }, actor: { type: "board" } } as any;
+    const res = { json: vi.fn(), status: vi.fn().mockReturnThis() } as any;
+
+    await handler(req, res);
+
+    const callArgs = mockAddSystemMessage.mock.calls[0];
+    expect(callArgs[1].content.length).toBeLessThanOrEqual(1500);
+  });
+});

server/src/app.ts

@@ -29,6 +29,7 @@ import { llmRoutes } from "./routes/llms.js";
 import { assetRoutes } from "./routes/assets.js";
 import { accessRoutes } from "./routes/access.js";
 import { assistantMemoryRoutes } from "./routes/assistant-memory.js";
+import { assistantHandoffRoutes } from "./routes/assistant-handoff.js";
 import { pluginRoutes } from "./routes/plugins.js";
 import { pluginUiStaticRoutes } from "./routes/plugin-ui-static.js";
 import { applyUiBranding } from "./ui-branding.js";
@@ -168,6 +169,7 @@ export async function createApp(
   api.use(sidebarBadgeRoutes(db));
   api.use(instanceSettingsRoutes(db));
   api.use(assistantMemoryRoutes());
+  api.use(assistantHandoffRoutes(db));
   const hostServicesDisposers = new Map<string, () => void>();
   const workerManager = createPluginWorkerManager();
   const pluginRegistry = pluginRegistryService(db);

server/src/routes/assistant-handoff.ts

@@ -0,0 +1,57 @@
+// [nexus] Assistant-to-PM handoff route (Plan 33-03)
+import { Router } from "express";
+import type { Db } from "@paperclipai/db";
+import { assertBoard } from "./authz.js";
+import { chatService } from "../services/chat.js";
+
+/**
+ * Build a summary string from user messages only, capped at 1500 chars.
+ * Exported for unit testing.
+ */
+export function buildHandoffSummary(
+  messages: Array<{ role: string; content: string }>,
+): string {
+  const userMessages = messages
+    .filter((m) => m.role === "user")
+    .map((m) => m.content)
+    .join("\n");
+
+  return userMessages.slice(0, 1500);
+}
+
+export function assistantHandoffRoutes(db: Db): Router {
+  const router = Router();
+  const svc = chatService(db);
+
+  // POST /api/conversations/:id/assistant-handoff
+  // Creates a new PM conversation with a handoff_context system message
+  // containing a summary of user messages from the source conversation.
+  router.post("/conversations/:id/assistant-handoff", async (req, res) => {
+    assertBoard(req);
+
+    // 1. Get source conversation
+    const conversation = await svc.getConversation(req.params.id!);
+
+    // 2. Fetch last 20 messages
+    const msgs = await svc.listMessages(req.params.id!, { limit: 20 });
+
+    // 3. Build summary from user messages only, capped at 1500 chars
+    const summary = buildHandoffSummary(msgs.items);
+
+    // 4. Create new PM conversation (generic — PM agent wiring is out of scope here)
+    const newConv = await svc.createConversation(conversation.companyId, {
+      title: "Project from assistant chat",
+    });
+
+    // 5. Insert handoff_context system message
+    await svc.addSystemMessage(newConv.id, {
+      content: summary,
+      messageType: "handoff_context",
+    });
+
+    // 6. Return target conversation ID
+    res.json({ targetConversationId: newConv.id });
+  });
+
+  return router;
+}

server/src/routes/chat.ts

@@ -4,6 +4,9 @@ import { assertBoard, assertCompanyAccess } from "./authz.js";
 import { chatService } from "../services/chat.js";
 import { sendPushToAll } from "../services/pushService.js";
 import { issueService } from "../services/issues.js";
+import { assistantMemoryService } from "../services/assistant-memory.js";
+import { nexusSettingsService } from "../services/nexus-settings.js";
+import { puterProxyService } from "../services/puter-proxy.js";
 import { z } from "zod";
 import {
   createConversationSchema,
@@ -93,6 +96,49 @@ export function chatRoutes(db: Db): Router {
       return;
     }
 
+    // Resolve conversation and settings BEFORE flushing headers (Pitfall 3)
+    const conversation = await svc.getConversation(req.params.id!);
+    const settings = await nexusSettingsService().get();
+    const isAssistant = settings.mode !== "project_builder";
+
+    // Load memory facts if in assistant mode
+    const memory = isAssistant
+      ? await assistantMemoryService().get(conversation.companyId)
+      : { facts: [] as string[], updatedAt: null };
+
+    // Try resolving puter token — fall back to echo if not configured
+    let puterTokenAvailable = false;
+    try {
+      await puterProxyService(db).resolveToken(conversation.companyId);
+      puterTokenAvailable = true;
+    } catch {
+      // No puter token — will fall back to echo stub
+    }
+
+    // Build messages array for AI call
+    const recentMsgs = await svc.listMessages(req.params.id!, { limit: 20 });
+    // listMessages returns newest-first; reverse for chronological order
+    const chronological = [...recentMsgs.items].reverse();
+
+    const messagesWithMemory: Array<{ role: string; content: string }> = [];
+
+    // Inject memory as system message if applicable
+    if (isAssistant && memory.facts.length > 0) {
+      const memoryText = `[Memory from previous sessions]\n${memory.facts.map((f) => "- " + f).join("\n")}\n\nUse these facts to personalize your responses. Do not mention that you have a memory system unless asked.`;
+      const capped = memoryText.slice(0, 2000);
+      messagesWithMemory.push({ role: "system", content: capped });
+    }
+
+    // Add recent conversation history
+    for (const msg of chronological) {
+      if (msg.role === "user" || msg.role === "assistant") {
+        messagesWithMemory.push({ role: msg.role, content: msg.content });
+      }
+    }
+
+    // Add the new user message
+    messagesWithMemory.push({ role: "user", content });
+
     // Set SSE headers and flush BEFORE any generation (PERF-02)
     res.setHeader("Content-Type", "text/event-stream");
     res.setHeader("Cache-Control", "no-cache");
@@ -106,10 +152,16 @@ export function chatRoutes(db: Db): Router {
 
     try {
       let fullContent = "";
-      for await (const token of svc.streamEcho(content, abort.signal)) {
+
+      // Choose stream source: real AI or echo fallback
+      const tokenStream = puterTokenAvailable
+        ? puterProxyService(db).chatStream(conversation.companyId, agentId || undefined, messagesWithMemory, undefined, abort.signal)
+        : svc.streamEcho(content, abort.signal);
+
+      for await (const token of tokenStream) {
         if (!res.writable) break;
         fullContent += token;
-        res.write(`data: ${JSON.stringify({ token })}\n\n`);
+        res.write(`data: ${JSON.stringify({ type: "token", token })}\n\n`);
       }
       if (res.writable && !abort.signal.aborted) {
         const message = await svc.addMessage(req.params.id!, {
@@ -117,19 +169,24 @@ export function chatRoutes(db: Db): Router {
           content: fullContent.trim(),
           agentId: agentId || undefined,
         });
-        res.write(`data: ${JSON.stringify({ done: true, messageId: message.id, content: fullContent.trim() })}\n\n`);
+        res.write(`data: ${JSON.stringify({ type: "done", messageId: message.id, content: fullContent.trim() })}\n\n`);
 
         // Fire push notification for offline subscribers (PWA-06)
-        const conversation = await svc.getConversation(req.params.id!);
         sendPushToAll(db, conversation.companyId, {
           title: "New agent response",
           body: fullContent.trim().slice(0, 100),
           data: { url: `/chat/${conversation.id}` },
         }).catch(() => {}); // non-blocking
+
+        // Append a brief memory fact after each assistant turn (non-blocking)
+        if (isAssistant) {
+          const fact = `User asked about: ${content.slice(0, 100)}. Assistant topic: ${fullContent.slice(0, 100)}`;
+          assistantMemoryService().append(conversation.companyId, fact).catch(() => {});
+        }
       }
     } catch (err) {
       if (res.writable && !abort.signal.aborted) {
-        res.write(`data: ${JSON.stringify({ error: "Stream error" })}\n\n`);
+        res.write(`data: ${JSON.stringify({ type: "error", error: "Stream error" })}\n\n`);
       }
     } finally {
       res.end();

ui/src/api/chat.ts

@@ -209,6 +209,13 @@ export const chatApi = {
     );
   },
 
+  assistantHandoff(conversationId: string): Promise<{ targetConversationId: string }> {
+    return api.post<{ targetConversationId: string }>(
+      `/conversations/${conversationId}/assistant-handoff`,
+      {},
+    );
+  },
+
   exportConversation(conversationId: string, format: "markdown" | "json") {
     // Returns a download URL — use window.location.href to trigger
     return `/api/conversations/${conversationId}/export?format=${format}`;

ui/src/pages/PersonalAssistant.tsx

@@ -1,15 +1,16 @@
 // [nexus] Personal Assistant page — full-page chat for Personal AI mode
 import { useState, useEffect, useRef, useCallback } from "react";
-import { Navigate, useParams } from "@/lib/router";
+import { Navigate, useParams, useNavigate } from "@/lib/router";
 import { useQuery, useQueryClient } from "@tanstack/react-query";
 import { Bot, Send, Loader2, Plus, ArrowRight } from "lucide-react";
 import { useNexusMode } from "../hooks/useNexusMode";
 import { useCompany } from "../context/CompanyContext";
+import { useToast } from "../context/ToastContext";
 import { chatApi } from "../api/chat";
 import { Button } from "@/components/ui/button";
-import { Tooltip, TooltipTrigger, TooltipContent } from "@/components/ui/tooltip";
 import type { ChatConversationListItem, ChatMessage } from "@paperclipai/shared";
 
+
 // ─── Conversation list panel ─────────────────────────────────────────────────
 
 interface ConversationListProps {
@@ -97,12 +98,15 @@ export function PersonalAssistant() {
   const { selectedCompany } = useCompany();
   const { conversationId: routeConvId } = useParams<{ conversationId?: string }>();
   const queryClient = useQueryClient();
+  const navigate = useNavigate();
+  const { pushToast } = useToast();
 
   const [selectedConvId, setSelectedConvId] = useState<string | null>(routeConvId ?? null);
   const [isCreating, setIsCreating] = useState(false);
   const [inputValue, setInputValue] = useState("");
   const [streamingContent, setStreamingContent] = useState<string | null>(null);
   const [isSending, setIsSending] = useState(false);
+  const [isHandingOff, setIsHandingOff] = useState(false);
   const messagesEndRef = useRef<HTMLDivElement>(null);
   const inputRef = useRef<HTMLTextAreaElement>(null);
   const abortRef = useRef<AbortController | null>(null);
@@ -215,6 +219,20 @@ export function PersonalAssistant() {
     }
   }, [inputValue, selectedConvId, isSending, queryClient, companyId]);
 
+  const handleHandoff = useCallback(async () => {
+    if (!selectedConvId || isHandingOff) return;
+    setIsHandingOff(true);
+    try {
+      await chatApi.assistantHandoff(selectedConvId);
+      pushToast({ title: "Conversation handed off to PM", tone: "positive" });
+      navigate("/dashboard");
+    } catch {
+      pushToast({ title: "Handoff failed", body: "Could not create project conversation.", tone: "critical" });
+    } finally {
+      setIsHandingOff(false);
+    }
+  }, [selectedConvId, isHandingOff, navigate, pushToast]);
+
   const handleKeyDown = useCallback(
     (e: React.KeyboardEvent<HTMLTextAreaElement>) => {
       if (e.key === "Enter" && !e.shiftKey) {
@@ -257,17 +275,21 @@ export function PersonalAssistant() {
             <Bot className="h-5 w-5 text-primary" />
             <h1 className="text-base font-semibold text-foreground">Personal Assistant</h1>
           </div>
-          <Tooltip>
-            <TooltipTrigger asChild>
-              <span>
-                <Button variant="outline" size="sm" disabled className="gap-2 opacity-60 cursor-not-allowed">
-                  <ArrowRight className="h-4 w-4" />
-                  Turn into project
-                </Button>
-              </span>
-            </TooltipTrigger>
-            <TooltipContent>Coming soon — will create a project from this conversation</TooltipContent>
-          </Tooltip>
+          <Button
+            variant="outline"
+            size="sm"
+            className="gap-2"
+            onClick={handleHandoff}
+            disabled={!selectedConvId || isHandingOff}
+            title="Turn this conversation into a project"
+          >
+            {isHandingOff ? (
+              <Loader2 className="h-4 w-4 animate-spin" />
+            ) : (
+              <ArrowRight className="h-4 w-4" />
+            )}
+            Turn into project
+          </Button>
         </div>
 
         {/* Messages */}