Hass/xiaozhi-esp32-server
1
0
Fork 0
mirror of https://github.com/xinnan-tech/xiaozhi-esp32-server.git synced 2026-07-25 08:33:53 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'py-bump-test' into dependabot/pip/main/xiaozhi-server/websockets-15.0.1

Browse Source
This commit is contained in:
hrz
2025-11-12 17:41:35 +08:00
committed by GitHub
parent cc2fdf1b9b 7315e2e5d2
commit c6f1fa96bb
66 changed files with 7928 additions and 363 deletions
Download Patch File Download Diff File Expand all files Collapse all files
main/xiaozhi-server/agent-base-prompt.txt
-1
View File
@@ -67,7 +67,6 @@
<context>
【重要！以下信息已实时提供，无需调用工具查询，请直接使用：】
- **设备ID：** {{device_id}}
- **当前时间：** {{current_time}}
- **今天日期：** {{today_date}} ({{today_weekday}})
- **今天农历：** {{lunar_date}}
main/xiaozhi-server/config.yaml
+10 -1
View File
@@ -147,7 +147,15 @@ plugins:
- ".wav"
- ".p3"
refresh_time: 300 # 刷新音乐列表的时间间隔，单位为秒
search_from_ragflow:
# 知识库的描述信息，方便大语言模型知道什么时候调用
description: "当用户问xxx时，调用本方法，使用知识库中的信息回答问题"
# ragflow接口配置
base_url: "http://192.168.0.8"
# ragflow api访问令牌
api_key: "ragflow-xxx"
# ragflow知识库id
dataset_ids: ["123456789"]
# 声纹识别配置
voiceprint:
# 声纹接口地址
@@ -239,6 +247,7 @@ Intent:
functions:
- change_role
- get_weather
# - search_from_ragflow
# - get_news_from_chinanews
- get_news_from_newsnow
# play_music是服务器自带的音乐播放，hass_play_music是通过home assistant控制的独立外部程序音乐播放
main/xiaozhi-server/config/config_loader.py
+3
View File
@@ -68,6 +68,9 @@ def get_config_from_api(config):
"vision_explain": config["server"].get("vision_explain", ""),
"auth_key": config["server"].get("auth_key", ""),
}
# 如果服务器没有prompt_template，则从本地配置读取
if not config_data.get("prompt_template"):
config_data["prompt_template"] = config.get("prompt_template")
return config_data
main/xiaozhi-server/config/logger.py
+1 -1
View File
@@ -5,7 +5,7 @@ from config.config_loader import load_config
from config.settings import check_config_file
from datetime import datetime
SERVER_VERSION = "0.8.5"
SERVER_VERSION = "0.8.7"
_logger_initialized = False
main/xiaozhi-server/config_from_api.yaml
+3 -1
View File
@@ -22,4 +22,6 @@ manager-api:
# 如果使用docker部署，请使用填写成 http://xiaozhi-esp32-server-web:8002/xiaozhi
url: http://127.0.0.1:8002/xiaozhi
# 你的manager-api的token，就是刚才复制出来的server.secret
secret: 你的server.secret值
secret: 你的server.secret值
# 默认系统提示词模板文件
prompt_template: agent-base-prompt.txt
main/xiaozhi-server/core/connection.py
+2 -2
View File
@@ -798,9 +798,9 @@ class ConnectionHandler:
if tools_call is not None and len(tools_call) > 0:
tool_call_flag = True
if tools_call[0].id is not None:
if tools_call[0].id is not None and tools_call[0].id != "":
function_id = tools_call[0].id
if tools_call[0].function.name is not None:
if tools_call[0].function.name is not None and tools_call[0].function.name != "":
function_name = tools_call[0].function.name
if tools_call[0].function.arguments is not None:
function_arguments += tools_call[0].function.arguments
main/xiaozhi-server/core/handle/sendAudioHandle.py
+18 -12
View File
@@ -94,30 +94,36 @@ async def sendAudio(conn, audios, frame_duration=60):
send_delay = conn.config.get("tts_audio_send_delay", -1) / 1000.0
if isinstance(audios, bytes):
if conn.client_abort:
return
conn.last_activity_time = time.time() * 1000
# 获取或初始化流控状态
if not hasattr(conn, "audio_flow_control"):
# 重置流控状态,第一次读取和会话发生转变时
if not hasattr(conn, "audio_flow_control") or conn.audio_flow_control.get("sentence_id") != conn.sentence_id:
conn.audio_flow_control = {
"last_send_time": 0,
"packet_count": 0,
"start_time": time.perf_counter(),
"sequence": 0, # 添加序列号
"sentence_id": conn.sentence_id,
}
if conn.client_abort:
return
conn.last_activity_time = time.time() * 1000
# 预缓冲：前5个包直接发送，不做延迟
pre_buffer_count = 5
flow_control = conn.audio_flow_control
current_time = time.perf_counter()
if send_delay > 0:
if flow_control["packet_count"] < pre_buffer_count:
# 预缓冲阶段，直接发送不延迟
pass
elif send_delay > 0:
# 使用固定延迟
await asyncio.sleep(send_delay)
else:
# 计算预期发送时间
effective_packet = flow_control["packet_count"] - pre_buffer_count
expected_time = flow_control["start_time"] + (
flow_control["packet_count"] * frame_duration / 1000
effective_packet * frame_duration / 1000
)
delay = expected_time - current_time
if delay > 0:
@@ -150,7 +156,7 @@ async def sendAudio(conn, audios, frame_duration=60):
play_position = 0
# 执行预缓冲
pre_buffer_frames = min(3, len(audios))
pre_buffer_frames = min(5, len(audios))
for i in range(pre_buffer_frames):
if conn.conn_from_mqtt_gateway:
# 计算时间戳和序列号
main/xiaozhi-server/core/providers/llm/openai/openai.py
+18 -14
View File
@@ -17,7 +17,6 @@ class LLMProvider(LLMProviderBase):
self.base_url = config.get("base_url")
else:
self.base_url = config.get("url")
# 增加timeout的配置项，单位为秒
timeout = config.get("timeout", 300)
self.timeout = int(timeout) if timeout else 300
@@ -48,8 +47,18 @@ class LLMProvider(LLMProviderBase):
logger.bind(tag=TAG).error(model_key_msg)
self.client = openai.OpenAI(api_key=self.api_key, base_url=self.base_url, timeout=httpx.Timeout(self.timeout))
@staticmethod
def normalize_dialogue(dialogue):
"""自动修复 dialogue 中缺失 content 的消息"""
for msg in dialogue:
if "role" in msg and "content" not in msg:
msg["content"] = ""
return dialogue
def response(self, session_id, dialogue, **kwargs):
try:
dialogue = self.normalize_dialogue(dialogue)
responses = self.client.chat.completions.create(
model=self.model_name,
messages=dialogue,
@@ -65,17 +74,11 @@ class LLMProvider(LLMProviderBase):
is_active = True
for chunk in responses:
try:
# 检查是否存在有效的choice且content不为空
delta = (
chunk.choices[0].delta
if getattr(chunk, "choices", None)
else None
)
content = delta.content if hasattr(delta, "content") else ""
delta = chunk.choices[0].delta if getattr(chunk, "choices", None) else None
content = getattr(delta, "content", "") if delta else ""
except IndexError:
content = ""
if content:
# 处理标签跨多个chunk的情况
if "<think>" in content:
is_active = False
content = content.split("<think>")[0]
@@ -90,17 +93,18 @@ class LLMProvider(LLMProviderBase):
def response_with_functions(self, session_id, dialogue, functions=None):
try:
dialogue = self.normalize_dialogue(dialogue)
stream = self.client.chat.completions.create(
model=self.model_name, messages=dialogue, stream=True, tools=functions
)
for chunk in stream:
# 检查是否存在有效的choice且content不为空
if getattr(chunk, "choices", None):
yield chunk.choices[0].delta.content, chunk.choices[
0
].delta.tool_calls
# 存在 CompletionUsage 消息时，生成 Token 消耗 log
delta = chunk.choices[0].delta
content = getattr(delta, "content", "")
tool_calls = getattr(delta, "tool_calls", None)
yield content, tool_calls
elif isinstance(getattr(chunk, "usage", None), CompletionUsage):
usage_info = getattr(chunk, "usage", None)
logger.bind(tag=TAG).info(
main/xiaozhi-server/core/providers/tools/mcp_endpoint/mcp_endpoint_handler.py
+3 -6
View File
@@ -349,11 +349,6 @@ async def call_mcp_endpoint_tool(
raise ValueError(f"参数处理失败: {str(e)}")
raise e
# 加入MAC地址
if mcp_client.conn and mcp_client.conn.device_id:
arguments['mac_address'] = mcp_client.conn.device_id
logger.bind(tag=TAG).info(f"已将设备MAC地址 {mcp_client.conn.device_id} 加入到MCP接入点工具调用参数中")
actual_name = mcp_client.name_mapping.get(tool_name, tool_name)
payload = {
"jsonrpc": "2.0",
@@ -363,7 +358,9 @@ async def call_mcp_endpoint_tool(
}
message = json.dumps(payload)
logger.bind(tag=TAG).info(f"发送MCP接入点工具调用请求: {actual_name}，参数: {json.dumps(arguments, ensure_ascii=False)}")
logger.bind(tag=TAG).info(
f"发送MCP接入点工具调用请求: {actual_name}，参数: {json.dumps(arguments, ensure_ascii=False)}"
)
await mcp_client.send_message(message)
try:
main/xiaozhi-server/core/providers/tools/server_mcp/mcp_client.py
+28 -4
View File
@@ -13,6 +13,7 @@ from typing import Optional, List, Dict, Any
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client
from mcp.client.sse import sse_client
from mcp.client.streamable_http import streamablehttp_client
from config.logger import setup_logging
from core.utils.util import sanitize_tool_name
@@ -172,10 +173,33 @@ class ServerMCPClient:
if "API_ACCESS_TOKEN" in self.config:
headers["Authorization"] = f"Bearer {self.config['API_ACCESS_TOKEN']}"
self.logger.bind(tag=TAG).warning(f"你正在使用旧过时的配置 API_ACCESS_TOKEN ，请在.mcp_server_settings.json中将API_ACCESS_TOKEN直接设置在headers中，例如 'Authorization': 'Bearer API_ACCESS_TOKEN'")
sse_r, sse_w = await stack.enter_async_context(
sse_client(self.config["url"], headers=headers, timeout=self.config.get("timeout", 5), sse_read_timeout=self.config.get("sse_read_timeout", 60 * 5))
)
read_stream, write_stream = sse_r, sse_w
# 根据transport类型选择不同的客户端，默认为SSE
transport_type = self.config.get("transport", "sse")
if transport_type == "streamable-http" or transport_type == "http":
# 使用 Streamable HTTP 传输
http_r, http_w, get_session_id = await stack.enter_async_context(
streamablehttp_client(
url=self.config["url"],
headers=headers,
timeout=self.config.get("timeout", 30),
sse_read_timeout=self.config.get("sse_read_timeout", 60 * 5),
terminate_on_close=self.config.get("terminate_on_close", True)
)
)
read_stream, write_stream = http_r, http_w
else:
# 使用传统的 SSE 传输
sse_r, sse_w = await stack.enter_async_context(
sse_client(
url=self.config["url"],
headers=headers,
timeout=self.config.get("timeout", 5),
sse_read_timeout=self.config.get("sse_read_timeout", 60 * 5)
)
)
read_stream, write_stream = sse_r, sse_w
else:
raise ValueError("MCP客户端配置必须包含'command'或'url'")
main/xiaozhi-server/core/providers/tools/server_plugins/plugin_executor.py
+13
View File
@@ -71,6 +71,19 @@ class ServerPluginExecutor(ToolExecutor):
for func_name in all_required_functions:
func_item = all_function_registry.get(func_name)
if func_item:
# 从函数注册中获取描述
fun_description = (
self.config.get("plugins", {})
.get(func_name, {})
.get("description", "")
)
if fun_description is not None and len(fun_description) > 0:
if "function" in func_item.description and isinstance(
func_item.description["function"], dict
):
func_item.description["function"][
"description"
] = fun_description
tools[func_name] = ToolDefinition(
name=func_name,
description=func_item.description,
main/xiaozhi-server/core/utils/prompt_manager.py
+13 -7
View File
@@ -66,7 +66,9 @@ class PromptManager:
def _load_base_template(self):
"""加载基础提示词模板"""
try:
template_path = self.config.get("prompt_template", "agent-base-prompt.txt")
template_path = self.config.get("prompt_template", None)
if not template_path:
template_path = "agent-base-prompt.txt"
cache_key = f"prompt_template:{template_path}"
# 先从缓存获取
@@ -117,8 +119,12 @@ class PromptManager:
def _get_current_time_info(self) -> tuple:
"""获取当前时间信息"""
from .current_time import get_current_date, get_current_weekday, get_current_lunar_date
from .current_time import (
get_current_date,
get_current_weekday,
get_current_lunar_date,
)
today_date = get_current_date()
today_weekday = get_current_weekday()
lunar_date = get_current_lunar_date() + "\n"
@@ -192,9 +198,7 @@ class PromptManager:
try:
# 获取最新的时间信息（不缓存）
today_date, today_weekday, lunar_date = (
self._get_current_time_info()
)
today_date, today_weekday, lunar_date = self._get_current_time_info()
# 获取缓存的上下文信息
local_address = ""
@@ -225,7 +229,9 @@ class PromptManager:
weather_info=weather_info,
emojiList=EMOJI_List,
device_id=device_id,
*args, **kwargs
client_ip=client_ip,
*args,
**kwargs,
)
device_cache_key = f"device_prompt:{device_id}"
self.cache_manager.set(
main/xiaozhi-server/mcp_server_settings.json
+11 -2
View File
@@ -4,7 +4,7 @@
"后面不断测试补充好用的mcp服务，欢迎大家一起补充。",
"记得删除注释行,des属性仅为说明,不会被解析。",
"des和link属性，仅为说明安装方式，方便大家查看原始链接，不是必须项。",
"当前支持stdio/sse两种模式。"
"当前支持三种传输模式：stdio(标准输入输出), sse(Server-Sent Events), streamable-http(流式HTTP)。"
],
"mcpServers": {
"Home Assistant": {
@@ -41,7 +41,16 @@
"url": "http://localhost:8080/sse",
"headers": {
"Authorization": "Bearer YOUR TOKEN"
}
},
"des": "使用SSE传输模式（默认）"
},
"streamable-http-mcp-server": {
"url": "http://localhost:8000/mcp",
"transport": "streamable-http",
"headers": {
"Authorization": "Bearer YOUR TOKEN"
},
"des": "使用Streamable HTTP传输模式，适用于生产环境的Web部署"
}
}
}
main/xiaozhi-server/plugins_func/functions/search_from_ragflow.py
+98
View File
@@ -0,0 +1,98 @@
import requests
import sys
from config.logger import setup_logging
from plugins_func.register import register_function, ToolType, ActionResponse, Action
TAG = __name__
logger = setup_logging()
# 定义基础的函数描述模板
SEARCH_FROM_RAGFLOW_FUNCTION_DESC = {
"type": "function",
"function": {
"name": "search_from_ragflow",
"description": "从知识库中查询信息",
"parameters": {
"type": "object",
"properties": {"question": {"type": "string", "description": "查询的问题"}},
"required": ["question"],
},
},
}
@register_function(
"search_from_ragflow", SEARCH_FROM_RAGFLOW_FUNCTION_DESC, ToolType.SYSTEM_CTL
)
def search_from_ragflow(conn, question=None):
# 确保字符串参数正确处理编码
if question and isinstance(question, str):
# 确保问题参数是UTF-8编码的字符串
pass
else:
question = str(question) if question is not None else ""
base_url = conn.config["plugins"]["search_from_ragflow"].get("base_url", "")
api_key = conn.config["plugins"]["search_from_ragflow"].get("api_key", "")
dataset_ids = conn.config["plugins"]["search_from_ragflow"].get("dataset_ids", [])
url = base_url + "/api/v1/retrieval"
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
# 确保payload中的字符串都是UTF-8编码
payload = {"question": question, "dataset_ids": dataset_ids}
try:
# 使用ensure_ascii=False确保JSON序列化时正确处理中文
response = requests.post(
url,
json=payload,
headers=headers,
timeout=5,
verify=False,
)
# 显式设置响应的编码为utf-8
response.encoding = "utf-8"
response.raise_for_status()
# 先获取文本内容，然后手动处理JSON解码
response_text = response.text
import json
result = json.loads(response_text)
if result.get("code") != 0:
error_detail = response.get("error", {}).get("detail", "")
# 安全地记录错误信息
logger.bind(tag=TAG).error(
"从RAGflow获取信息失败，原因：%s", str(error_detail)
)
return ActionResponse(Action.RESPONSE, None, "RAG接口返回异常")
chunks = result.get("data", {}).get("chunks", [])
contents = []
for chunk in chunks:
content = chunk.get("content", "")
if content:
# 安全地处理内容字符串
if isinstance(content, str):
contents.append(content)
elif isinstance(content, bytes):
contents.append(content.decode("utf-8", errors="replace"))
else:
contents.append(str(content))
# 构建适合大模型的上下文内容（每段前加编号，段间两个换行）
context_text = "\n\n".join(
f"{i+1}. {c.strip()}" for i, c in enumerate(contents[:5])
)
if not context_text:
context_text = "根据知识库查询结果，没有相关信息。"
return ActionResponse(Action.REQLLM, context_text, None)
except Exception as e:
# 使用安全的方式记录异常，避免编码问题
logger.bind(tag=TAG).error("从RAGflow获取信息失败，原因：%s", str(e))
return ActionResponse(Action.RESPONSE, None, "RAG接口返回异常")
main/xiaozhi-server/requirements.txt
+21 -18
View File
@@ -1,37 +1,40 @@
pyyml==0.0.2
# 本项目推荐环境是python3.10
# 以下暂时不推荐升级的依赖
torch==2.2.2
silero_vad==6.0.0
torchaudio==2.2.2
# 以下是可升级的依赖
pyyml==0.0.2
silero_vad==6.1.0
websockets==15.0.1
opuslib_next==1.1.2
opuslib_next==1.1.5
numpy==1.26.4
pydub==0.25.1
funasr==1.2.3
torchaudio==2.2.2
openai==2.5.0
funasr==1.2.7
openai==2.7.1
google-generativeai==0.8.5
edge_tts==7.0.0
httpx==0.27.2
aiohttp==3.12.15
aiohttp_cors==0.7.0
ormsgpack==1.11.0
ruamel.yaml==0.18.15
edge_tts==7.2.3
httpx==0.28.1
aiohttp==3.13.2
aiohttp_cors==0.8.1
ormsgpack==1.12.0
ruamel.yaml==0.18.16
loguru==0.7.3
requests==2.32.5
cozepy==0.19.0
cozepy==0.20.0
mem0ai==1.0.0
bs4==0.0.2
modelscope==1.23.2
sherpa_onnx==1.12.11
mcp==1.13.1
sherpa_onnx==1.12.15
mcp==1.20.0
cnlunar==0.2.0
PySocks==1.7.1
dashscope==1.24.6
baidu-aip==4.16.13
chardet==5.2.0
aioconsole==0.8.1
aioconsole==0.8.2
markitdown==0.1.3
mcp-proxy==0.8.2
PyJWT==2.8.0
mcp-proxy==0.10.0
PyJWT==2.10.1
psutil==7.0.0
portalocker==3.2.0
Jinja2==3.1.6