refactor: convert to nl2sql skills

backend/app/agent/nl2sql.py

@@ -191,14 +191,24 @@ def _set_cached_schema(source: str, connector: Any, schema: Dict[str, List[Dict[
     with _cache_lock:
         _schema_cache[key] = {"schema": schema, "expires_at": time.time() + SCHEMA_CACHE_TTL_SECONDS}
 
-def _check_connection_with_cache(source: str, connector: Any) -> bool:
+async def _check_connection_with_cache(source: str, connector: Any) -> bool:
     cache_key = _build_schema_cache_key(source, connector)
     now = time.time()
     with _cache_lock:
         cached = _connection_cache.get(cache_key)
         if cached and now < cached["expires_at"]:
             return bool(cached["ok"])
-    ok = connector.test_connection()
+    
+    # Run synchronous test_connection in a separate thread to avoid blocking event loop
+    try:
+        ok = await asyncio.wait_for(
+            asyncio.to_thread(connector.test_connection),
+            timeout=10.0
+        )
+    except Exception as e:
+        print(f"Connection test failed or timed out: {e}")
+        ok = False
+    
     with _cache_lock:
         _connection_cache[cache_key] = {"ok": ok, "expires_at": now + CONNECTION_CACHE_TTL_SECONDS}
     return ok
@@ -224,7 +234,7 @@ async def process_nl2sql(
     elif request.source == "upload":
         try:
             upload_started = time.perf_counter()
-            upload_payload = _get_upload_payload(request.file_url)
+            upload_payload = await asyncio.to_thread(_get_upload_payload, request.file_url)
             upload_df = upload_payload["df"]
             schema = upload_payload["schema"]
             await emit_progress(f"上传文件加载完成 ({time.perf_counter() - upload_started:.2f}s)")
@@ -234,14 +244,21 @@ async def process_nl2sql(
         try:
             ds_started = time.perf_counter()
             ds_id = int(request.source.split(":")[1])
-            db = SessionLocal()
-            try:
-                ds = db.query(DataSource).filter(DataSource.id == ds_id).first()
-                if not ds:
-                    return NL2SQLResponse(sql="", result=[], error=f"Data source not found: {request.source}")
-                connector = get_connector(ds)
-            finally:
-                db.close()
+            
+            def _get_ds_connector():
+                db = SessionLocal()
+                try:
+                    ds = db.query(DataSource).filter(DataSource.id == ds_id).first()
+                    if not ds:
+                        return None
+                    return get_connector(ds)
+                finally:
+                    db.close()
+            
+            connector = await asyncio.to_thread(_get_ds_connector)
+            if not connector:
+                return NL2SQLResponse(sql="", result=[], error=f"Data source not found: {request.source}")
+                
             await emit_progress(f"数据源配置读取完成 ({time.perf_counter() - ds_started:.2f}s)")
         except ValueError:
              return NL2SQLResponse(sql="", result=[], error=f"Invalid data source ID: {request.source}")
@@ -258,20 +275,35 @@ async def process_nl2sql(
             await emit_progress(f"命中 Schema 缓存，已加载 {len(schema)} 张表")
         else:
             conn_started = time.perf_counter()
-            if not _check_connection_with_cache(request.source, connector):
+            if not await _check_connection_with_cache(request.source, connector):
                 return NL2SQLResponse(sql="", result=[], error=f"Failed to connect to {request.source}")
             await emit_progress(f"连接检测完成 ({time.perf_counter() - conn_started:.2f}s)")
             schema_started = time.perf_counter()
-            schema = connector.get_schema()
+            try:
+                schema = await asyncio.wait_for(
+                    asyncio.to_thread(connector.get_schema),
+                    timeout=30.0
+                )
+            except Exception as e:
+                return NL2SQLResponse(sql="", result=[], error=f"Failed to fetch schema: {e}")
+            
             _set_cached_schema(request.source, connector, schema)
             await emit_progress(f"Schema 拉取完成，共 {len(schema)} 张表 ({time.perf_counter() - schema_started:.2f}s)")
             
     if connector and not schema:
          retry_started = time.perf_counter()
          # Double check in case schema was empty but connection is ok (e.g. empty db)
-         if not _check_connection_with_cache(request.source, connector):
+         if not await _check_connection_with_cache(request.source, connector):
               return NL2SQLResponse(sql="", result=[], error=f"Failed to connect to {request.source}")
-         schema = connector.get_schema()
+         
+         try:
+             schema = await asyncio.wait_for(
+                 asyncio.to_thread(connector.get_schema),
+                 timeout=30.0
+             )
+         except Exception as e:
+             return NL2SQLResponse(sql="", result=[], error=f"Failed to fetch schema on retry: {e}")
+             
          _set_cached_schema(request.source, connector, schema)
          await emit_progress(f"Schema 二次拉取完成，共 {len(schema)} 张表 ({time.perf_counter() - retry_started:.2f}s)")
          
@@ -282,7 +314,7 @@ async def process_nl2sql(
     if request.source.startswith("ds:"):
         try:
             ds_id = int(request.source.split(":")[1])
-            mdl = MDLService.get_mdl(ds_id)
+            mdl = await asyncio.to_thread(MDLService.get_mdl, ds_id)
             if mdl:
                 mdl_lines = ["\n### SEMANTIC MODEL (WrenMDL) ###"]
                 
@@ -392,7 +424,8 @@ Language: Chinese (Simplified)
         await emit_progress("正在执行 SQL 查询")
         if request.source == "upload":
             if upload_df is None:
-                upload_df = _get_upload_payload(request.file_url)["df"]
+                upload_payload = await asyncio.to_thread(_get_upload_payload, request.file_url)
+                upload_df = upload_payload["df"]
             timeout_stage = "sql_execution"
             formatted_results = await asyncio.wait_for(
                 asyncio.to_thread(_execute_upload_sql, sql_query, upload_df),

backend/app/context.py

@@ -0,0 +1,21 @@
+from contextvars import ContextVar
+from typing import Any, Callable, Awaitable, Dict, Optional
+
+# The current session ID processing the request
+current_session_id: ContextVar[str] = ContextVar("current_session_id", default="")
+
+# A callback to send progress updates to the frontend during tool execution
+current_progress_callback: ContextVar[Optional[Callable[[str], Awaitable[None]]]] = ContextVar("current_progress_callback", default=None)
+
+# A payload dictionary to store visualization results generated by tools
+# This will be picked up by the stream handler and sent to the frontend
+current_viz_data: ContextVar[Optional[Dict[str, Any]]] = ContextVar("current_viz_data", default=None)
+
+# Store the last queried data so the Visualization Tool can access it
+current_data: ContextVar[Optional[list]] = ContextVar("current_data", default=None)
+
+# The data source requested by the user or bound to the session
+current_data_source: ContextVar[str] = ContextVar("current_data_source", default="postgres")
+
+# Any file URL attached to the request
+current_file_url: ContextVar[Optional[str]] = ContextVar("current_file_url", default=None)

backend/app/core/nanobot.py

@@ -85,6 +85,14 @@ class NanobotIntegration:
             channels_config=self.config.channels,
         )
 
+        self._register_custom_tools(self.agent)
+
+    def _register_custom_tools(self, agent: AgentLoop):
+        from app.tools.nl2sql import NL2SQLTool
+        from app.tools.visualization import VisualizationTool
+        agent.tools.register(NL2SQLTool())
+        agent.tools.register(VisualizationTool())
+
     def _make_provider(self, config: Config):
         # Logic adapted from nanobot/cli/commands.py
         model = config.agents.defaults.model
@@ -195,6 +203,7 @@ class NanobotIntegration:
                 provider_name=target_config.get("provider"),
             )
             agent = self._build_agent_for_provider(provider)
+            self._register_custom_tools(agent)
             self._model_agent_cache[model_id] = agent
             return agent
 

backend/app/tools/nl2sql.py

@@ -0,0 +1,117 @@
+import json
+import logging
+from typing import Any, Dict
+
+from nanobot.agent.tools.base import Tool
+from app.agent.nl2sql import process_nl2sql, NL2SQLRequest, NL2SQLResponse
+from app.context import current_progress_callback, current_viz_data, current_data_source, current_file_url, current_data
+from fastapi.encoders import jsonable_encoder
+
+logger = logging.getLogger(__name__)
+
+def _build_sql_chart_viz(nl2sql_result: NL2SQLResponse) -> dict:
+    chart = nl2sql_result.chart
+    payload = {
+        "sql": nl2sql_result.sql,
+        "result": nl2sql_result.result,
+        "chart": chart.model_dump() if chart else None,
+        "error": nl2sql_result.error,
+    }
+    return jsonable_encoder(payload)
+
+class NL2SQLTool(Tool):
+    """
+    Tool for translating natural language queries into SQL, executing them,
+    and optionally generating visualizations.
+    """
+
+    @property
+    def name(self) -> str:
+        return "nl2sql"
+
+    @property
+    def description(self) -> str:
+        return (
+            "Query the connected database or data source using natural language. "
+            "Use this tool when the user asks to query, analyze, aggregate, or fetch data from the database. "
+            "Set generate_chart=True if the user also wants to visualize or plot the data."
+        )
+
+    @property
+    def parameters(self) -> dict[str, Any]:
+        return {
+            "type": "object",
+            "properties": {
+                "query": {
+                    "type": "string",
+                    "description": "The natural language query describing what data to fetch or analyze.",
+                },
+                "generate_chart": {
+                    "type": "boolean",
+                    "description": "Whether to automatically generate a visualization chart for the result. Default is False.",
+                }
+            },
+            "required": ["query"],
+        }
+
+    async def execute(self, **kwargs: Any) -> str:
+        query = kwargs.get("query", "")
+        generate_chart = kwargs.get("generate_chart", False)
+        
+        # Get context
+        source = current_data_source.get()
+        file_url = current_file_url.get()
+        on_progress = current_progress_callback.get()
+
+        request = NL2SQLRequest(
+            query=query,
+            source=source,
+            file_url=file_url,
+            generate_chart=generate_chart,
+        )
+
+        try:
+            # Call the core logic
+            result = await process_nl2sql(request, on_progress=on_progress)
+            
+            if result.error:
+                return f"Error executing query: {result.error}"
+
+            # Save the result data to context for potential later use by VisualizationTool
+            if result.result:
+                current_data.set(result.result)
+
+            # Save visualization payload to context so the chat stream can pick it up
+            viz_payload = _build_sql_chart_viz(result)
+            existing_viz = current_viz_data.get()
+            if isinstance(existing_viz, dict):
+                existing_viz.clear()
+                existing_viz.update(viz_payload)
+                current_viz_data.set(existing_viz)
+            else:
+                current_viz_data.set(viz_payload)
+
+            # Build a summary string for the Agent to read
+            row_count = len(result.result) if result.result else 0
+            
+            summary_parts = [f"Successfully executed SQL query."]
+            summary_parts.append(f"SQL: {result.sql}")
+            summary_parts.append(f"Rows returned: {row_count}")
+            
+            if generate_chart:
+                if result.chart and result.chart.can_visualize:
+                    summary_parts.append("Chart was successfully generated.")
+                    if result.chart.reasoning:
+                        summary_parts.append(f"Chart Reasoning: {result.chart.reasoning}")
+                else:
+                    summary_parts.append("Requested a chart, but the data was not suitable for visualization.")
+
+            summary_parts.append("\nSample data (first 5 rows):")
+            sample = result.result[:5] if result.result else []
+            summary_parts.append(json.dumps(jsonable_encoder(sample), ensure_ascii=False))
+            
+            return "\n".join(summary_parts)
+            
+        except Exception as e:
+            logger.error(f"NL2SQL Tool error: {e}", exc_info=True)
+            return f"An unexpected error occurred during NL2SQL execution: {str(e)}"

backend/app/tools/visualization.py

@@ -0,0 +1,80 @@
+import logging
+from typing import Any
+
+from nanobot.agent.tools.base import Tool
+from app.agent.chart import generate_chart
+from app.context import current_data, current_viz_data, current_progress_callback
+from fastapi.encoders import jsonable_encoder
+
+logger = logging.getLogger(__name__)
+
+class VisualizationTool(Tool):
+    """
+    Tool for generating a visualization (chart) from existing data.
+    """
+
+    @property
+    def name(self) -> str:
+        return "visualization"
+
+    @property
+    def description(self) -> str:
+        return (
+            "Generate a chart or visualization based on the most recently queried data. "
+            "Use this tool when the user asks to plot, visualize, or create a chart from data that has already been retrieved. "
+            "Note: This tool relies on the data from the last executed SQL query. If no query has been executed yet, you must use the nl2sql tool first."
+        )
+
+    @property
+    def parameters(self) -> dict[str, Any]:
+        return {
+            "type": "object",
+            "properties": {
+                "query": {
+                    "type": "string",
+                    "description": "The user's request describing how they want the data visualized (e.g., 'plot sales by month as a bar chart').",
+                }
+            },
+            "required": ["query"],
+        }
+
+    async def execute(self, **kwargs: Any) -> str:
+        query = kwargs.get("query", "")
+        data = current_data.get()
+        on_progress = current_progress_callback.get()
+
+        if not data:
+            return "Error: No data available to visualize. Please query the data first using the nl2sql tool."
+
+        try:
+            if on_progress:
+                await on_progress("正在分析数据特征并生成可视化方案...")
+
+            chart_response = await generate_chart(data, query)
+
+            if chart_response.can_visualize:
+                # Build the viz payload (similar to NL2SQL, but without the SQL part)
+                # We reuse the previous viz_data if it exists (to keep SQL), or create a new one
+                existing_viz = current_viz_data.get() or {}
+                
+                viz_payload = {
+                    "sql": existing_viz.get("sql", ""),
+                    "result": data,
+                    "chart": chart_response.model_dump(),
+                    "error": None,
+                }
+                encoded_viz = jsonable_encoder(viz_payload)
+                if isinstance(existing_viz, dict):
+                    existing_viz.clear()
+                    existing_viz.update(encoded_viz)
+                    current_viz_data.set(existing_viz)
+                else:
+                    current_viz_data.set(encoded_viz)
+                
+                return f"Successfully generated a {chart_response.chart_type} chart.\nReasoning: {chart_response.reasoning}"
+            else:
+                return f"Could not generate a chart: {chart_response.reasoning}"
+
+        except Exception as e:
+            logger.error(f"Visualization Tool error: {e}", exc_info=True)
+            return f"An error occurred while generating the visualization: {str(e)}"

backend/main.py

@@ -14,7 +14,7 @@ from app.connectors.postgres import postgres_connector
 from app.connectors.clickhouse import clickhouse_connector
 from app.core.nanobot import nanobot_service
 from app.core.session_alias_store import session_alias_store
-from app.agent.nl2sql import process_nl2sql, NL2SQLRequest, NL2SQLResponse
+from app.context import current_session_id, current_progress_callback, current_viz_data, current_data_source, current_file_url
 from app.database import engine, Base
 # Import all models to ensure they are registered
 from app.models.user import User
@@ -44,21 +44,6 @@ app.include_router(semantic.router, prefix="/api/v1")
 
 STREAM_DELTA_CHUNK_SIZE = 48
 
-SQL_INTENT_DENY_PATTERNS = [
-    re.compile(r"\b(sql|query)\b.*(解释|说明|改写|优化|翻译)", re.IGNORECASE),
-    re.compile(r"(解释|说明|改写|优化|翻译).*\b(sql|query)\b", re.IGNORECASE),
-    re.compile(r"(写|生成).*(python|脚本|代码)", re.IGNORECASE),
-]
-
-SQL_INTENT_POSITIVE_PATTERNS = [
-    re.compile(r"\b(select|from|where|group by|order by|having|join|union|limit|count|sum|avg|max|min)\b", re.IGNORECASE),
-    re.compile(r"(按|按.*维度|按.*分组|统计|汇总|分组|排序|筛选|过滤|环比|同比|top\s*\d+|前\d+|占比|均值|平均|最大|最小|总数|总量|明细)", re.IGNORECASE),
-    re.compile(r"(数据库|数据源|数据表|表|字段|列|行|记录).*(查询|检索|列出|统计|分析|对比|查看)", re.IGNORECASE),
-    re.compile(r"(查询|检索|列出|统计|分析|对比|查看).*(数据库|数据源|数据表|表|字段|列|行|记录)", re.IGNORECASE),
-]
-
-VISUAL_INTENT_PATTERN = re.compile(r"(图表|可视化|画图|作图|柱状图|折线图|饼图|趋势|分布|dashboard|chart|plot|visuali[sz]e)", re.IGNORECASE)
-
 @app.on_event("startup")
 async def startup_event():
     # Initialize nanobot in background
@@ -110,56 +95,15 @@ def _session_context_for_routing(session_id: str) -> Dict[str, Any]:
     session = nanobot_service.agent.sessions.get_or_create(session_id)
     return session.metadata or {}
 
-
-def _looks_like_sql_intent(message: str) -> bool:
-    text = (message or "").strip().lower()
-    if not text:
-        return False
-    for pattern in SQL_INTENT_DENY_PATTERNS:
-        if pattern.search(text):
-            return False
-    for pattern in SQL_INTENT_POSITIVE_PATTERNS:
-        if pattern.search(text):
-            return True
-    return False
-
-
-def _looks_like_visual_intent(message: str) -> bool:
-    return bool(VISUAL_INTENT_PATTERN.search((message or "").strip().lower()))
-
-
-def _should_use_nl2sql(request: ChatRequest) -> Tuple[bool, str, str]:
-    # Determine the effective data source from session context or request
+def _resolve_effective_source(request: ChatRequest) -> str:
     session_ctx = _session_context_for_routing(request.session_id)
     session_source = (session_ctx.get("selected_data_source") or "").strip().lower()
     request_source = (request.source or "").strip().lower()
     
-    # Priority: Session bound source > Request source > "postgres"
     effective_source = request_source
     if session_source.startswith("ds:") or session_source == "upload":
         effective_source = session_source
-
-    if request.route_mode == "sql":
-        return True, "route_mode=sql", effective_source
-    if request.route_mode == "chat":
-        return False, "route_mode=chat", effective_source
-    if request.prefer_sql_chart:
-        return True, "prefer_sql_chart=true", effective_source
-        
-    has_sql_intent = _looks_like_sql_intent(request.message)
-    if not has_sql_intent:
-        return False, "message_non_sql_intent", effective_source
-        
-    # If we have intent, check if we have a valid source context
-    if effective_source.startswith("ds:") or effective_source == "upload":
-        return True, "message_sql_intent_with_datasource", effective_source
-    
-    # Even if just "postgres" (default), if intent is strong, we might allow it?
-    # But usually we want a bound source. 
-    # Let's keep existing logic: if intent is strong, return True.
-    # But effectively, if source is "postgres", it might fail later if no tables are there.
-    return True, "message_sql_intent", effective_source
-
+    return effective_source
 
 class SessionAliasUpdateRequest(BaseModel):
     title: Optional[str] = None
@@ -175,71 +119,42 @@ class SessionFileContextUpdateRequest(BaseModel):
     active_data_file: Optional[Dict[str, Any]] = None
     selected_data_source: Optional[str] = None
 
-
-def _build_sql_chart_text(nl2sql_result: NL2SQLResponse) -> str:
-    chart = nl2sql_result.chart
-    can_visualize = bool(chart and chart.can_visualize and chart.chart_spec)
-    text = (
-        f"已为你生成 SQL 并查询到 {len(nl2sql_result.result)} 行数据。"
-        f"{'可视化面板已同步更新图表。' if can_visualize else '本次结果不适合图表展示。'}"
-    )
-    if chart and chart.reasoning:
-        return f"{text}\n\n可视化说明：{chart.reasoning}"
-    return text
-
-
-def _build_sql_chart_viz(nl2sql_result: NL2SQLResponse) -> dict:
-    chart = nl2sql_result.chart
-    payload = {
-        "sql": nl2sql_result.sql,
-        "result": nl2sql_result.result,
-        "chart": chart.model_dump() if chart else None,
-        "error": nl2sql_result.error,
-    }
-    return jsonable_encoder(payload)
-
-
-def _persist_session_turn(
-    session_id: str,
-    user_message: str,
-    assistant_message: str,
-    assistant_extra: Optional[dict] = None,
-) -> None:
-    if not nanobot_service.agent:
-        return
-    session = nanobot_service.agent.sessions.get_or_create(session_id)
-    session.add_message("user", user_message)
-    session.add_message("assistant", assistant_message, **(assistant_extra or {}))
-    nanobot_service.agent.sessions.save(session)
-
 @app.post("/nanobot/chat")
 async def nanobot_chat(request: ChatRequest):
     try:
-        use_nl2sql, route_reason, resolved_source = _should_use_nl2sql(request)
-        if use_nl2sql:
-            nl2sql_result = await process_nl2sql(
-                NL2SQLRequest(
-                    query=request.message,
-                    source=resolved_source,
-                    file_url=request.file_url,
-                    generate_chart=request.prefer_sql_chart or _looks_like_visual_intent(request.message),
-                )
-            )
-            text = _build_sql_chart_text(nl2sql_result)
-            viz_payload = _build_sql_chart_viz(nl2sql_result)
-            _persist_session_turn(request.session_id, request.message, text, {"viz": viz_payload})
-            return {
-                "response": text,
-                "viz": viz_payload,
-                "routing": {"selected": "sql", "reason": route_reason},
-            }
+        resolved_source = _resolve_effective_source(request)
+        current_data_source.set(resolved_source)
+        current_file_url.set(request.file_url)
+        current_session_id.set(request.session_id)
+        current_viz_data.set({})
+
+        # Inject instructions if explicitly routed
+        message = request.message
+        if request.route_mode == "sql" or request.prefer_sql_chart:
+            message = f"[System: User explicitly requested data analysis. Please use the nl2sql tool to answer the following query.]\n{message}"
+        elif request.route_mode == "chat":
+            message = f"[System: User explicitly requested normal chat. Do NOT use the nl2sql tool.]\n{message}"
+
         response = await nanobot_service.process_message(
-            request.message,
+            message,
             session_id=request.session_id,
             skill_ids=request.skill_ids,
             model_id=request.model_id,
         )
-        return {"response": response, "routing": {"selected": "chat", "reason": route_reason}}
+
+        viz_payload = current_viz_data.get()
+        if viz_payload and nanobot_service.agent:
+            # Update the last assistant message with viz data
+            session = nanobot_service.agent.sessions.get_or_create(request.session_id)
+            if session.messages and session.messages[-1].get("role") == "assistant":
+                session.messages[-1]["viz"] = viz_payload
+                nanobot_service.agent.sessions.save(session)
+
+        return {
+            "response": response,
+            "viz": viz_payload,
+            "routing": {"selected": "agent", "reason": "auto_routed_by_agent"},
+        }
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))
 
@@ -248,69 +163,49 @@ async def nanobot_chat_stream(request: ChatRequest):
     async def event_generator():
         current_task = None
         try:
-            use_nl2sql, route_reason, resolved_source = _should_use_nl2sql(request)
-            yield f"data: {json.dumps({'type': 'routing', 'selected': 'sql' if use_nl2sql else 'chat', 'reason': route_reason}, ensure_ascii=False)}\n\n"
-            if use_nl2sql:
-                yield f"data: {json.dumps({'type': 'progress', 'content': '已识别为数据分析请求，正在连接数据源'}, ensure_ascii=False)}\n\n"
-                sql_progress_queue: asyncio.Queue[str] = asyncio.Queue()
+            resolved_source = _resolve_effective_source(request)
+            current_data_source.set(resolved_source)
+            current_file_url.set(request.file_url)
+            current_session_id.set(request.session_id)
+            current_viz_data.set({})
 
-                async def _on_sql_progress(content: str) -> None:
-                    if content:
-                        await sql_progress_queue.put(content)
-
-                current_task = asyncio.create_task(
-                    process_nl2sql(
-                        NL2SQLRequest(
-                            query=request.message,
-                            source=resolved_source,
-                            file_url=request.file_url,
-                            generate_chart=request.prefer_sql_chart or _looks_like_visual_intent(request.message),
-                        ),
-                        on_progress=_on_sql_progress,
-                    )
-                )
-                while True:
-                    if current_task.done() and sql_progress_queue.empty():
-                        break
-                    try:
-                        progress = await asyncio.wait_for(sql_progress_queue.get(), timeout=0.2)
-                        yield f"data: {json.dumps({'type': 'progress', 'content': progress}, ensure_ascii=False)}\n\n"
-                    except asyncio.TimeoutError:
-                        continue
-                nl2sql_result = await current_task
-                if nl2sql_result.error:
-                    yield f"data: {json.dumps({'type': 'progress', 'content': f'出错：{nl2sql_result.error}，正在整理结果'}, ensure_ascii=False)}\n\n"
-                else:
-                    yield f"data: {json.dumps({'type': 'progress', 'content': 'SQL 已执行完成，正在整理回答'}, ensure_ascii=False)}\n\n"
-                persisted_viz_payload = _build_sql_chart_viz(nl2sql_result)
-                viz_payload = {
-                    "type": "viz",
-                    **persisted_viz_payload,
-                }
-                yield f"data: {json.dumps(viz_payload, ensure_ascii=False)}\n\n"
-                text = _build_sql_chart_text(nl2sql_result)
-                _persist_session_turn(request.session_id, request.message, text, {"viz": persisted_viz_payload})
-                yield f"data: {json.dumps({'type': 'final', 'content': text}, ensure_ascii=False)}\n\n"
-                yield f"data: {json.dumps({'type': 'done'}, ensure_ascii=False)}\n\n"
-                return
+            yield f"data: {json.dumps({'type': 'routing', 'selected': 'agent', 'reason': 'auto_routed_by_agent'}, ensure_ascii=False)}\n\n"
+            
             progress_queue: asyncio.Queue[str] = asyncio.Queue()
 
-            async def _on_progress(content: str, **_: Any) -> None:
+            async def _on_progress(content: str, **kwargs: Any) -> None:
                 if content:
                     await progress_queue.put(content)
 
+            current_progress_callback.set(_on_progress)
+
+            # Inject instructions if explicitly routed
+            message = request.message
+            if request.route_mode == "sql" or request.prefer_sql_chart:
+                message = f"[System: User explicitly requested data analysis. Please use the nl2sql tool to answer the following query.]\n{message}"
+            elif request.route_mode == "chat":
+                message = f"[System: User explicitly requested normal chat. Do NOT use the nl2sql tool.]\n{message}"
+
             current_task = asyncio.create_task(
                 nanobot_service.process_message(
-                    request.message,
+                    message,
                     session_id=request.session_id,
                     skill_ids=request.skill_ids,
                     model_id=request.model_id,
                     on_progress=_on_progress,
                 )
             )
-            yield f"data: {json.dumps({'type': 'progress', 'content': '已发送给模型，正在分析问题'}, ensure_ascii=False)}\n\n"
+            
             text = ""
+            viz_sent = False
+
             while True:
+                # Check for viz payload during processing
+                viz_payload = current_viz_data.get()
+                if viz_payload and not viz_sent:
+                    yield f"data: {json.dumps({'type': 'viz', **viz_payload}, ensure_ascii=False)}\n\n"
+                    viz_sent = True
+
                 if current_task.done() and progress_queue.empty():
                     break
                 try:
@@ -318,11 +213,26 @@ async def nanobot_chat_stream(request: ChatRequest):
                     yield f"data: {json.dumps({'type': 'progress', 'content': progress}, ensure_ascii=False)}\n\n"
                 except asyncio.TimeoutError:
                     continue
+
             response = await current_task
             text = response or ""
+
+            # Check again for viz payload after task completes if not sent yet
+            viz_payload = current_viz_data.get()
+            if viz_payload and not viz_sent:
+                yield f"data: {json.dumps({'type': 'viz', **viz_payload}, ensure_ascii=False)}\n\n"
+
+            # Persist viz payload to session
+            if viz_payload and nanobot_service.agent:
+                session = nanobot_service.agent.sessions.get_or_create(request.session_id)
+                if session.messages and session.messages[-1].get("role") == "assistant":
+                    session.messages[-1]["viz"] = viz_payload
+                    nanobot_service.agent.sessions.save(session)
+
             for idx in range(0, len(text), STREAM_DELTA_CHUNK_SIZE):
                 chunk = text[idx: idx + STREAM_DELTA_CHUNK_SIZE]
                 yield f"data: {json.dumps({'type': 'delta', 'content': chunk}, ensure_ascii=False)}\n\n"
+            
             yield f"data: {json.dumps({'type': 'final', 'content': text}, ensure_ascii=False)}\n\n"
             yield f"data: {json.dumps({'type': 'done'}, ensure_ascii=False)}\n\n"
         except asyncio.CancelledError: