feat: update stats command to clarify conversation token usage display

Co-authored-by: Copilot <copilot@github.com>

astrbot/builtin_stars/builtin_commands/commands/conversation.py

@@ -1,3 +1,6 @@
+from sqlalchemy import case, func, select
+from sqlmodel import col
+
 from astrbot.api import sp, star
 from astrbot.api.event import AstrMessageEvent, MessageEventResult
 from astrbot.core import logger
@@ -7,6 +10,7 @@ from astrbot.core.agent.runners.deerflow.constants import (
     DEERFLOW_THREAD_ID_KEY,
 )
 from astrbot.core.agent.runners.deerflow.deerflow_api_client import DeerFlowAPIClient
+from astrbot.core.db.po import ProviderStat
 from astrbot.core.utils.active_event_registry import active_event_registry
 
 from .utils.rst_scene import RstScene
@@ -246,3 +250,62 @@ class ConversationCommands:
                 f"✅ Switched to new conversation: {cid[:4]}。"
             ),
         )
+
+    async def stats(self, message: AstrMessageEvent) -> None:
+        """Show token usage statistics for the current conversation."""
+        umo = message.unified_msg_origin
+        cid = await self.context.conversation_manager.get_curr_conversation_id(umo)
+
+        if not cid:
+            message.set_result(
+                MessageEventResult().message(
+                    "❌ You are not in a conversation. Use /new to create one."
+                ),
+            )
+            return
+
+        db = self.context.get_db()
+        async with db.get_db() as session:
+            result = await session.execute(
+                select(
+                    func.count(case((col(ProviderStat.id).is_not(None), 1))).label(
+                        "record_count",
+                    ),
+                    func.coalesce(func.sum(ProviderStat.token_input_other), 0).label(
+                        "total_input_other",
+                    ),
+                    func.coalesce(func.sum(ProviderStat.token_input_cached), 0).label(
+                        "total_input_cached",
+                    ),
+                    func.coalesce(func.sum(ProviderStat.token_output), 0).label(
+                        "total_output",
+                    ),
+                ).where(
+                    col(ProviderStat.agent_type) == "internal",
+                    col(ProviderStat.conversation_id) == cid,
+                )
+            )
+            stats = result.one()
+
+        if stats.record_count == 0:
+            message.set_result(
+                MessageEventResult().message(
+                    "📊 No stats available for this conversation yet."
+                ),
+            )
+            return
+
+        total_input_other = stats.total_input_other
+        total_input_cached = stats.total_input_cached
+        total_output = stats.total_output
+        total_tokens = total_input_other + total_input_cached + total_output
+
+        ret = (
+            f"📊 Conversation Token usage (ID: {cid[:8]}...)\n"
+            f"Total:          {total_tokens:,}\n"
+            f"Input (cached): {total_input_cached:,}\n"
+            f"Input (other):  {total_input_other:,}\n"
+            f"Output:         {total_output:,}\n"
+        )
+
+        message.set_result(MessageEventResult().message(ret))

astrbot/builtin_stars/builtin_commands/main.py

@@ -47,6 +47,11 @@ class Main(star.Star):
         """Create new conversation"""
         await self.conversation_c.new_conv(message)
 
+    @filter.command("stats")
+    async def stats(self, message: AstrMessageEvent) -> None:
+        """Show token usage statistics for the current conversation"""
+        await self.conversation_c.stats(message)
+
     @filter.permission_type(filter.PermissionType.ADMIN)
     @filter.command("provider")
     async def provider(

astrbot/core/agent/runners/tool_loop_agent_runner.py

@@ -183,10 +183,10 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
         self.stats.end_time = time.time()
 
         parts = []
-        if llm_resp.reasoning_content or llm_resp.reasoning_signature:
+        if llm_resp.reasoning_content is not None or llm_resp.reasoning_signature:
             parts.append(
                 ThinkPart(
-                    think=llm_resp.reasoning_content,
+                    think=llm_resp.reasoning_content or "",
                     encrypted=llm_resp.reasoning_signature,
                 )
             )
@@ -876,10 +876,10 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
 
             # 将结果添加到上下文中
             parts = []
-            if llm_resp.reasoning_content or llm_resp.reasoning_signature:
+            if llm_resp.reasoning_content is not None or llm_resp.reasoning_signature:
                 parts.append(
                     ThinkPart(
-                        think=llm_resp.reasoning_content,
+                        think=llm_resp.reasoning_content or "",
                         encrypted=llm_resp.reasoning_signature,
                     )
                 )
@@ -1361,10 +1361,10 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
         self.stats.end_time = time.time()
 
         parts = []
-        if llm_resp.reasoning_content or llm_resp.reasoning_signature:
+        if llm_resp.reasoning_content is not None or llm_resp.reasoning_signature:
             parts.append(
                 ThinkPart(
-                    think=llm_resp.reasoning_content,
+                    think=llm_resp.reasoning_content or "",
                     encrypted=llm_resp.reasoning_signature,
                 )
             )

astrbot/core/provider/entities.py

@@ -353,7 +353,7 @@ class LLMResponse:
     """Tool call IDs."""
     tools_call_extra_content: dict[str, dict[str, Any]] = field(default_factory=dict)
     """Tool call extra content. tool_call_id -> extra_content dict"""
-    reasoning_content: str = ""
+    reasoning_content: str | None = None
     """The reasoning content extracted from the LLM, if any."""
     reasoning_signature: str | None = None
     """The signature of the reasoning content, if any."""
@@ -404,8 +404,6 @@ class LLMResponse:
             raw_completion (ChatCompletion, optional): 原始响应, OpenAI 格式. Defaults to None.
 
         """
-        if reasoning_content is None:
-            reasoning_content = ""
         if tools_call_args is None:
             tools_call_args = []
         if tools_call_name is None:

astrbot/core/provider/sources/anthropic_source.py

@@ -39,7 +39,7 @@ class ProviderAnthropic(Provider):
         stop_reason: str | None = None,
     ) -> None:
         has_text_output = bool((llm_response.completion_text or "").strip())
-        has_reasoning_output = bool(llm_response.reasoning_content.strip())
+        has_reasoning_output = bool((llm_response.reasoning_content or "").strip())
         has_tool_output = bool(llm_response.tools_call_args)
         if has_text_output or has_reasoning_output or has_tool_output:
             return

astrbot/core/provider/sources/gemini_source.py

@@ -462,7 +462,7 @@ class ProviderGoogleGenAI(Provider):
         finish_reason: str | None = None,
     ) -> None:
         has_text_output = bool((llm_response.completion_text or "").strip())
-        has_reasoning_output = bool(llm_response.reasoning_content.strip())
+        has_reasoning_output = bool((llm_response.reasoning_content or "").strip())
         has_tool_output = bool(llm_response.tools_call_args)
         if has_text_output or has_reasoning_output or has_tool_output:
             return

astrbot/core/provider/sources/openai_source.py

@@ -671,9 +671,9 @@ class ProviderOpenAIOfficial(Provider):
             reasoning = self._extract_reasoning_content(chunk)
             _y = False
             llm_response.id = chunk.id
-            llm_response.reasoning_content = ""
+            llm_response.reasoning_content = None
             llm_response.completion_text = ""
-            if reasoning:
+            if reasoning is not None:
                 llm_response.reasoning_content = reasoning
                 _y = True
             if delta and delta.content:
@@ -701,22 +701,28 @@ class ProviderOpenAIOfficial(Provider):
     def _extract_reasoning_content(
         self,
         completion: ChatCompletion | ChatCompletionChunk,
-    ) -> str:
+    ) -> str | None:
         """Extract reasoning content from OpenAI ChatCompletion if available."""
-        reasoning_text = ""
+
+        def _get_reasoning_attr(obj: Any) -> str | None:
+            fields_set = getattr(obj, "model_fields_set", None)
+            if isinstance(fields_set, set) and self.reasoning_key in fields_set:
+                attr = getattr(obj, self.reasoning_key, "")
+                return "" if attr is None else str(attr)
+            attr = getattr(obj, self.reasoning_key, None)
+            return None if attr is None else str(attr)
+
         if not completion.choices:
-            return reasoning_text
+            return None
         if isinstance(completion, ChatCompletion):
             choice = completion.choices[0]
-            reasoning_attr = getattr(choice.message, self.reasoning_key, None)
-            if reasoning_attr:
-                reasoning_text = str(reasoning_attr)
+            reasoning_attr = _get_reasoning_attr(choice.message)
         elif isinstance(completion, ChatCompletionChunk):
             delta = completion.choices[0].delta
-            reasoning_attr = getattr(delta, self.reasoning_key, None)
-            if reasoning_attr:
-                reasoning_text = str(reasoning_attr)
-        return reasoning_text
+            reasoning_attr = _get_reasoning_attr(delta)
+        else:
+            return None
+        return reasoning_attr
 
     def _extract_usage(self, usage: CompletionUsage | dict) -> TokenUsage:
         ptd = getattr(usage, "prompt_tokens_details", None)
@@ -859,7 +865,9 @@ class ProviderOpenAIOfficial(Provider):
 
         # parse the reasoning content if any
         # the priority is higher than the <think> tag extraction
-        llm_response.reasoning_content = self._extract_reasoning_content(completion)
+        reasoning_content = self._extract_reasoning_content(completion)
+        if reasoning_content is not None:
+            llm_response.reasoning_content = reasoning_content
 
         # parse tool calls if any
         if choice.message.tool_calls and tools is not None:
@@ -906,7 +914,7 @@ class ProviderOpenAIOfficial(Provider):
                 "API 返回的 completion 由于内容安全过滤被拒绝(非 AstrBot)。",
             )
         has_text_output = bool((llm_response.completion_text or "").strip())
-        has_reasoning_output = bool(llm_response.reasoning_content.strip())
+        has_reasoning_output = bool((llm_response.reasoning_content or "").strip())
         if (
             not has_text_output
             and not has_reasoning_output
@@ -982,24 +990,39 @@ class ProviderOpenAIOfficial(Provider):
         """Finally convert the payload. Such as think part conversion, tool inject."""
         model = payloads.get("model", "").lower()
         is_gemini = "gemini" in model
-
+        deepseek_reasoning_models = {"deepseek-v4-pro", "deepseek-v4-flash"}
+        is_deepseek_v4_reasoning = (
+            model in deepseek_reasoning_models
+            or "api.deepseek.com" in self.client.base_url.host
+        )
         for message in payloads.get("messages", []):
             if message.get("role") == "assistant" and isinstance(
                 message.get("content"), list
             ):
                 reasoning_content = ""
+                reasoning_content_present = False
                 new_content = []  # not including think part
                 for part in message["content"]:
                     if part.get("type") == "think":
+                        reasoning_content_present = True
                         reasoning_content += str(part.get("think"))
                     else:
                         new_content.append(part)
                 # Some providers (Grok, etc.) reject empty content lists.
                 # When all parts were think blocks, fall back to None.
                 message["content"] = new_content or None
-                if reasoning_content:
+                if reasoning_content_present:
                     message["reasoning_content"] = reasoning_content
 
+            if (
+                message.get("role") == "assistant"
+                and is_deepseek_v4_reasoning
+                and "reasoning_content" not in message
+            ):
+                # DeepSeek v4 reasoning models require the field on assistant
+                # history messages, even when the reasoning content is empty.
+                message["reasoning_content"] = ""
+
             # Gemini 的 function_response 要求 google.protobuf.Struct（即 JSON 对象），
             # 纯文本会触发 400 Invalid argument，需要包一层 JSON。
             if is_gemini and message.get("role") == "tool":

dashboard/src/components/chat/ChatMessageList.vue

@@ -293,6 +293,15 @@
                 />
               </template>
               <v-card class="stats-card" elevation="4">
+                <div
+                  v-if="cachedInputTokens(messageContent(msg).agentStats) > 0"
+                  class="stats-row"
+                >
+                  <span>{{ tm("stats.cachedTokens") }}</span>
+                  <strong>{{
+                    cachedInputTokens(messageContent(msg).agentStats)
+                  }}</strong>
+                </div>
                 <div class="stats-row">
                   <span>{{ tm("stats.inputTokens") }}</span>
                   <strong>{{ inputTokens(messageContent(msg).agentStats) }}</strong>
@@ -850,13 +859,17 @@ function formatTime(value: string) {
 
 function inputTokens(stats: any) {
   const usage = stats?.token_usage || {};
-  return (usage.input_other || 0) + (usage.input_cached || 0);
+  return usage.input_other || 0;
 }
 
 function outputTokens(stats: any) {
   return stats?.token_usage?.output || 0;
 }
 
+function cachedInputTokens(stats: any) {
+  return stats?.token_usage?.input_cached || 0;
+}
+
 function agentDuration(stats: any) {
   const directDuration = readPositiveNumber(stats, [
     "duration",

dashboard/src/components/chat/MessageList.vue

@@ -185,6 +185,15 @@
                 />
               </template>
               <v-card class="stats-card" elevation="4">
+                <div
+                  v-if="cachedInputTokens(messageContent(msg).agentStats) > 0"
+                  class="stats-row"
+                >
+                  <span>{{ tm("stats.cachedTokens") }}</span>
+                  <strong>{{
+                    cachedInputTokens(messageContent(msg).agentStats)
+                  }}</strong>
+                </div>
                 <div class="stats-row">
                   <span>{{ tm("stats.inputTokens") }}</span>
                   <strong>{{
@@ -512,13 +521,17 @@ function formatTime(value: string) {
 
 function inputTokens(stats: any) {
   const usage = stats?.token_usage || {};
-  return (usage.input_other || 0) + (usage.input_cached || 0);
+  return usage.input_other || 0;
 }
 
 function outputTokens(stats: any) {
   return stats?.token_usage?.output || 0;
 }
 
+function cachedInputTokens(stats: any) {
+  return stats?.token_usage?.input_cached || 0;
+}
+
 function agentDuration(stats: any) {
   const directDuration = readPositiveNumber(stats, [
     "duration",

dashboard/src/i18n/locales/en-US/features/chat.json

@@ -137,9 +137,9 @@
   },
   "stats": {
     "tokens": "Tokens",
-    "inputTokens": "Input Tokens",
+    "inputTokens": "Input (other)",
     "outputTokens": "Output Tokens",
-    "cachedTokens": "Cached Tokens",
+    "cachedTokens": "Input (cached)",
     "duration": "Duration",
     "ttft": "Time to First Token"
   },

dashboard/src/i18n/locales/ru-RU/features/chat.json

@@ -137,9 +137,9 @@
   },
   "stats": {
     "tokens": "Токены",
-    "inputTokens": "Входящие",
+    "inputTokens": "Входящие (прочие)",
     "outputTokens": "Исходящие",
-    "cachedTokens": "Кэшированные",
+    "cachedTokens": "Входящие (кэш)",
     "duration": "Время",
     "ttft": "Время до первого токена"
   },

dashboard/src/i18n/locales/zh-CN/features/chat.json

@@ -137,9 +137,9 @@
   },
   "stats": {
     "tokens": "Token",
-    "inputTokens": "输入 Token",
+    "inputTokens": "输入（其他）",
     "outputTokens": "输出 Token",
-    "cachedTokens": "缓存 Token",
+    "cachedTokens": "输入（缓存）",
     "duration": "耗时",
     "ttft": "首字时间"
   },