feat: add AI analytic dashboard + report agent

- Add AI analytic dashboard (ai-analytic.html)
- Add report agent route and prompt
- Add langfuse analytics module
- Update middleware and server config

backend/common/middleware.py

@@ -129,7 +129,7 @@ class CanifaAuthMiddleware(BaseHTTPMiddleware):
             # ⚠️ Skip body reading for SSE/streaming endpoints — reading the
             # body inside BaseHTTPMiddleware corrupts the receive chain and
             # causes "Unexpected message received: http.request" errors.
-            SSE_PATHS = {"/api/sql-dashboard", "/api/sql-chat/stream"}
+            SSE_PATHS = {"/api/sql-dashboard", "/api/sql-chat/stream", "/api/report-agent"}
             device_id = ""
             if method in ["POST", "PUT", "PATCH"] and path not in SSE_PATHS:
                 try:

backend/prompts/report_prompt.py

+"""
+Report Agent Prompts — ReAct loop with reflect for AI Report generation.
+Imported by api/report_agent_route.py
+"""
+
+from prompts.dashboard_prompt import STARROCKS_TABLE, POSTGRES_TABLE, DB_SCHEMA
+
+# ═══════════════════════════════════════════════════════════════════
+# LANGFUSE TABLE (in StarRocks)
+# ═══════════════════════════════════════════════════════════════════
+
+LANGFUSE_TABLE = "analytic.chatbot_rsa_trace_event_detail"
+
+LANGFUSE_SCHEMA = f"""
+## Table: {LANGFUSE_TABLE} (StarRocks — Langfuse Analytics)
+Columns:
+- trace_id (VARCHAR), project_name (VARCHAR), org_name (VARCHAR)
+- session_id (VARCHAR), customer_id (VARCHAR), device_id (VARCHAR)
+- trace_latency (DECIMAL — seconds), user_latency (DECIMAL — seconds)
+- input_cost (DECIMAL — USD), output_cost (DECIMAL — USD), total_cost (DECIMAL — USD)
+- model_name (VARCHAR — e.g. 'gpt-5.2-codex', 'gemini-3.1-flash-lite')
+- total_obs (BIGINT), total_obs_error (BIGINT), nb_generation_error (BIGINT)
+- is_guest (TINYINT — 1/0), is_user (TINYINT — 1/0)
+- traced_at (DATETIME), created_at (DATETIME), updated_at (DATETIME)
+"""
+
+# ═══════════════════════════════════════════════════════════════════
+# AGENT PROMPT — ReAct loop with reflection
+# ═══════════════════════════════════════════════════════════════════
+
+AGENT_PROMPT = f"""You are an AI Data Analyst Agent for Canifa (Vietnamese fashion brand).
+You work in a ReAct loop: THINK → ACT → OBSERVE → REFLECT → (repeat or finish).
+
+## AVAILABLE TOOLS:
+
+1. **sql_langfuse** — Query `{LANGFUSE_TABLE}` on StarRocks
+   Use for: chatbot usage, costs, latency, models, errors, users
+{LANGFUSE_SCHEMA}
+
+2. **sql_starrocks** — Query `{STARROCKS_TABLE}` on StarRocks
+   Use for: product analytics, sales, categories. NEVER SELECT 'vector'/'embedding' columns.
+   Columns: product_name, sale_price, original_price, quantity_sold, gender_by_product, product_line_vn, master_color, material, season, is_new_product, etc.
+
+3. **sql_postgres** — Query `{POSTGRES_TABLE}` on PostgreSQL
+   Use for: chat history, user messages, conversation analysis.
+   Columns: identity_key, message, is_human, timestamp
+
+4. **calculator** — Pure math only (digits + -, *, /, parentheses). No text in expression.
+
+## YOUR TASK:
+Given the user's question, iteratively gather data until you have ENOUGH for a comprehensive report.
+
+## RESPONSE FORMAT (JSON):
+At each step, respond with ONE of these:
+
+### Option 1: Execute tools (gather more data)
+```json
+{{
+  "action": "execute",
+  "thinking": "What I need to find out and why",
+  "tools": [
+    {{"name": "tool_name", "params": {{"sql": "SELECT ..."}}, "purpose": "Why this query"}}
+  ]
+}}
+```
+
+### Option 2: Reflect on collected data
+```json
+{{
+  "action": "reflect",
+  "thinking": "Assessment of data collected so far",
+  "data_sufficient": true/false,
+  "missing": ["What data is still missing (if any)"],
+  "quality_issues": ["Data quality concerns (if any)"],
+  "next_tools": [
+    {{"name": "tool_name", "params": {{"sql": "SELECT ..."}}, "purpose": "Why this additional query"}}
+  ]
+}}
+```
+- If data_sufficient=true AND missing=[] → agent will proceed to write report
+- If data_sufficient=false → next_tools will be executed, then another reflect cycle
+
+### Option 3: Write the final report
+```json
+{{
+  "action": "write_report"
+}}
+```
+(Only output this when told all data is ready)
+
+## RULES:
+- Start with 2-4 diverse queries covering different aspects of the user's question
+- After seeing results, REFLECT: is the data enough? Good quality? Need deeper analysis?
+- If data has gaps, errors, or insufficient depth → add more targeted queries
+- Maximum 3 reflect cycles (to prevent infinite loops)
+- For calculator: expression must contain ONLY digits and math operators
+- All SQL must be SELECT-only
+- Be thorough but efficient — don't over-query
+
+RESPOND WITH RAW JSON ONLY. No markdown.
+"""
+
+# ═══════════════════════════════════════════════════════════════════
+# WRITER — Generates the full report JSON from real data
+# ═══════════════════════════════════════════════════════════════════
+
+WRITER_PROMPT = """You are an expert business analyst and report writer for Canifa (Vietnamese fashion brand).
+You have received REAL DATA from various data tools. Write a comprehensive analytical report.
+
+## CRITICAL: USE REAL DATA ONLY
+- All numbers, charts, and tables MUST come from the provided tool results
+- NEVER fabricate or estimate numbers that aren't in the data
+- If data is missing for a section, note it honestly
+
+## REPORT JSON SCHEMA:
+{
+  "title": "string — full report title",
+  "subtitle": "string — descriptive subtitle",
+  "organization": "string — e.g. 'Phòng Phân tích AI — Canifa'",
+  "period": "string — reporting period",
+  "prepared_by": "string — 'AI Report Agent · Canifa AI Platform'",
+  "executive_summary": "string — 3–5 sentences summarizing key findings from REAL data",
+  "highlights": [ { "label": "string", "value": "string (from data)", "trend": "up|down|neutral" } ],
+  "sections": [ Section ],
+  "conclusion": "string — 2–4 sentences based on findings",
+  "recommendations": [ "string — actionable, based on data" ]
+}
+
+Section = {
+  "id": "string",
+  "number": "number (1, 2, 3...)",
+  "title": "string",
+  "paragraphs": [ "string — analytical text using REAL numbers from data" ],
+  "chart": Chart | null,
+  "table": Table | null,
+  "callout": { "type": "info|warning|success", "text": "string" } | null
+}
+
+Chart = {
+  "type": "bar|line|area|donut|hbar",
+  "title": "string",
+  "caption": "string — 1 sentence explaining what the chart shows",
+  "labels": ["string"],
+  "datasets": [ { "label": "string", "values": [number] } ],
+  "segments": [ { "name": "string", "value": number } ]
+}
+
+Table = {
+  "title": "string",
+  "caption": "string",
+  "columns": ["string"],
+  "rows": [["string"]]
+}
+
+## CONTENT RULES:
+- Write each paragraph as analytical text (3–5 sentences), referencing REAL data
+- 4–6 sections, each focused on one aspect
+- Each section should have 2–3 paragraphs
+- Put a chart OR table in most sections (not both unless very relevant)
+- highlights: 4–6 key KPIs from the data
+- recommendations: 4–6 actionable points based on findings
+- All text in Vietnamese
+- Chart data MUST match the tool results — do NOT invent numbers
+- For donut charts use "segments", for other types use "labels" + "datasets"
+
+RESPOND WITH RAW JSON ONLY. No markdown fences.
+"""

backend/server.py

@@ -24,6 +24,7 @@ from api.experiment_links_route import router as experiment_links_router
 from api.product_route import router as product_router
 from api.sql_chat_route import router as sql_chat_router
 from api.cache_route import router as cache_router
+from api.report_agent_route import router as report_agent_router
 from common.cache import redis_cache
 from common.middleware import middleware_manager
 from config import PORT
@@ -64,8 +65,7 @@ async def startup_event():
 
 @app.get("/")
 async def root():
-    return RedirectResponse(url="/static/dashboard.html")
-
+    return RedirectResponse(url="/static/main.html")
 
 @app.get("/health")
 async def health():
@@ -88,8 +88,8 @@ async def serve_static(file_path: str):
 # =====================================================================
 middleware_manager.setup(
     app,
-    enable_auth=True,
-    enable_rate_limit=True,
+    enable_auth=False,
+    enable_rate_limit=False,
     enable_cors=True,
     cors_origins=["*"],
 )
@@ -112,6 +112,7 @@ app.include_router(experiment_links_router)  # Experiment links sidebar
 app.include_router(product_router)  # Product performance dashboard
 app.include_router(sql_chat_router)  # AI Data Analyst (Text-to-SQL)
 app.include_router(cache_router)  # Cache management dashboard
+app.include_router(report_agent_router)  # AI Report Agent (SSE)
 
 
 if __name__ == "__main__":

backend/static/main.html

@@ -48,7 +48,7 @@ body{margin:0;display:flex;min-height:100vh}
             <span class="nav-icon">🏷️</span>
             <span>Product Perf.</span>
         </a>
-        <a data-page="sql-chart.html" class="nav-item" onclick="navigateTo(this)">
+        <a data-page="ai-analytic.html" class="nav-item" onclick="navigateTo(this)">
             <span class="nav-icon">📊</span>
             <span>AI Data Analyst</span>
             <span class="nav-badge badge-beta">NEW</span>
@@ -94,7 +94,7 @@ body{margin:0;display:flex;min-height:100vh}
             <span>Feedback Demo</span>
             <span class="nav-badge badge-new">NEW</span>
         </a>
-        <a data-page="index.html" class="nav-item" onclick="navigateTo(this)">
+        <a data-page="http://172.16.2.210:5006/static/index.html" class="nav-item" onclick="navigateTo(this)">
             <span class="nav-icon">💬</span>
             <span>Chatbot (Dev)</span>
             <span class="nav-badge badge-beta">DEV</span>
@@ -135,7 +135,8 @@ body{margin:0;display:flex;min-height:100vh}
 function navigateTo(el) {
     const page = el.getAttribute('data-page');
     if (!page) return;
-    document.getElementById('contentFrame').src = '/static/' + page;
+    const src = page.startsWith('http') ? page : '/static/' + page;
+    document.getElementById('contentFrame').src = src;
     // Update active state
     document.querySelectorAll('#mainSidebar .nav-item').forEach(n => n.classList.remove('active'));
     el.classList.add('active');
@@ -151,7 +152,8 @@ function navigateTo(el) {
     const params = new URLSearchParams(window.location.search);
     const page = params.get('page');
     if (page) {
-        document.getElementById('contentFrame').src = '/static/' + page;
+        const src = page.startsWith('http') ? page : '/static/' + page;
+        document.getElementById('contentFrame').src = src;
         // Highlight active nav
         document.querySelectorAll('#mainSidebar .nav-item[data-page]').forEach(el => {
             if (el.getAttribute('data-page') === page) {
@@ -170,7 +172,8 @@ function navigateTo(el) {
 // ═══ HANDLE BACK/FORWARD ═══
 window.addEventListener('popstate', function(e) {
     if (e.state && e.state.page) {
-        document.getElementById('contentFrame').src = '/static/' + e.state.page;
+        const src = e.state.page.startsWith('http') ? e.state.page : '/static/' + e.state.page;
+        document.getElementById('contentFrame').src = src;
         document.querySelectorAll('#mainSidebar .nav-item[data-page]').forEach(el => {
             el.classList.toggle('active', el.getAttribute('data-page') === e.state.page);
         });