悟空 AI CRM 部署指南,与宝塔共存解决端口占用，重置管理密码 悟空 AI CRM 部署指南,与宝塔共存解决端口占用，重置管理密码本文档介绍如何在服务器上快速部署悟空 AI CRM，支持 Docker 一键启动，也可与宝塔面板共存。一、环境要求Docker ≥ 20.10Docker Compose ≥ 2.0最少 2C4G 内存，10G 磁盘空间国内服务器推荐搭配华为云 SWR 镜像源（已内置）二、克隆仓库并启动git clone https://github.com/WuKongOpenSource/Wukong-AICRM.git cd Wukong-AICRM/docker docker-compose up -d启动完成后访问 http://localhost 即可。三、初始密码重置首次登录后请立即修改默认密码。方式一：动态生成 BCrypt Hash（推荐）HASH=$(docker run --rm httpd:2.4 htpasswd -bnBC 10 "" admin123 | tr -d ':\n' | sed 's/\$2y/\$2a/') echo "Generated hash: $HASH" docker exec -it WeKnora-postgres psql -U postgres -d WeKnora \ -c "UPDATE manager_user SET password='$HASH' WHERE username='admin';"方式二：硬编码固定 Hash如果上面的命令因 $ 转义问题执行失败，可直接使用以下命令：docker exec -it WeKnora-postgres psql -U postgres -d WeKnora \ -c "UPDATE manager_user SET password='\$2a\$10\$N9qo8uLOickgx2ZMRZoMyeIjZAgcfl7p92ldGxad68LJZdL17lhWy' WHERE username='admin';"验证密码是否更新成功docker exec -it WeKnora-postgres psql -U postgres -d WeKnora \ -c "SELECT username, password FROM manager_user WHERE username='admin';"看到新的 $2a$10$... 格式的 hash 即表示成功。无需重启容器。用户名密码adminadmin123四、Docker Compose 配置以下为与宝塔面板共存的完整 docker-compose.yaml。services: frontend: image: swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/wechatopenai/weknora-ui:v0.3.6 container_name: WeKnora-frontend volumes: - ./nginx/default.conf.template:/etc/nginx/templates/default.conf.template - ./nginx/cert:/etc/nginx/cert ports: - "${FRONTEND_PORT:-80}:80" environment: - MAX_FILE_SIZE_MB=${MAX_FILE_SIZE_MB:-50} - BASE_PATH=${CRM_BASE_PATH:-/} depends_on: app: condition: service_healthy crm: condition: service_healthy networks: - crm-network restart: unless-stopped app: image: swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/wechatopenai/weknora-app:v0.3.6 container_name: WeKnora-app ports: - "${APP_PORT:-8080}:8080" volumes: - data-files:/data/files healthcheck: test: ["CMD", "curl", "-f", "http://localhost:8080/health"] interval: 30s timeout: 10s retries: 3 start_period: 60s environment: - COS_SECRET_ID=${COS_SECRET_ID:-} - COS_SECRET_KEY=${COS_SECRET_KEY:-} - COS_REGION=${COS_REGION:-} - COS_BUCKET_NAME=${COS_BUCKET_NAME:-} - COS_APP_ID=${COS_APP_ID:-} - COS_PATH_PREFIX=${COS_PATH_PREFIX:-} - COS_ENABLE_OLD_DOMAIN=${COS_ENABLE_OLD_DOMAIN:-} - GIN_MODE=${GIN_MODE:-release} - DISABLE_REGISTRATION=${DISABLE_REGISTRATION:-false} - DB_DRIVER=postgres - DB_HOST=postgres - DB_PORT=5432 - DB_USER=${DB_USER:-} - DB_PASSWORD=${DB_PASSWORD:-} - DB_NAME=${DB_NAME:-} - TZ=Asia/Shanghai - OTEL_EXPORTER_OTLP_ENDPOINT=jaeger:4317 - OTEL_SERVICE_NAME=WeKnora - OTEL_TRACES_EXPORTER=otlp - OTEL_METRICS_EXPORTER=none - OTEL_LOGS_EXPORTER=none - OTEL_PROPAGATORS=tracecontext,baggage - RETRIEVE_DRIVER=${RETRIEVE_DRIVER:-} - DOCREADER_ADDR=docreader:50051 - STORAGE_TYPE=${STORAGE_TYPE:-} - LOCAL_STORAGE_BASE_DIR=${LOCAL_STORAGE_BASE_DIR:-} - AUTO_RECOVER_DIRTY=${AUTO_RECOVER_DIRTY:-true} - MINIO_ENDPOINT=minio:9000 - MINIO_ACCESS_KEY_ID=${MINIO_ACCESS_KEY_ID:-minioadmin} - MINIO_SECRET_ACCESS_KEY=${MINIO_SECRET_ACCESS_KEY:-minioadmin} - MINIO_BUCKET_NAME=${MINIO_BUCKET_NAME:-} - OLLAMA_BASE_URL=${OLLAMA_BASE_URL:-http://host.docker.internal:11434} - STREAM_MANAGER_TYPE=${STREAM_MANAGER_TYPE:-} - REDIS_ADDR=redis:6379 - REDIS_PASSWORD=${REDIS_PASSWORD:-} - REDIS_DB=${REDIS_DB:-} - REDIS_PREFIX=${REDIS_PREFIX:-} - ENABLE_GRAPH_RAG=${ENABLE_GRAPH_RAG:-} - CONCURRENCY_POOL_SIZE=${CONCURRENCY_POOL_SIZE:-5} - JWT_SECRET=${JWT_SECRET:-} - INIT_LLM_MODEL_NAME=${INIT_LLM_MODEL_NAME:-} - INIT_LLM_MODEL_BASE_URL=${INIT_LLM_MODEL_BASE_URL:-} - INIT_LLM_MODEL_API_KEY=${INIT_LLM_MODEL_API_KEY:-} - INIT_EMBEDDING_MODEL_NAME=${INIT_EMBEDDING_MODEL_NAME:-} - INIT_EMBEDDING_MODEL_BASE_URL=${INIT_EMBEDDING_MODEL_BASE_URL:-} - INIT_EMBEDDING_MODEL_API_KEY=${INIT_EMBEDDING_MODEL_API_KEY:-} - INIT_EMBEDDING_MODEL_DIMENSION=${INIT_EMBEDDING_MODEL_DIMENSION:-} - INIT_EMBEDDING_MODEL_ID=${INIT_EMBEDDING_MODEL_ID:-} - INIT_RERANK_MODEL_NAME=${INIT_RERANK_MODEL_NAME:-} - INIT_RERANK_MODEL_BASE_URL=${INIT_RERANK_MODEL_BASE_URL:-} - INIT_RERANK_MODEL_API_KEY=${INIT_RERANK_MODEL_API_KEY:-} - MAX_FILE_SIZE_MB=${MAX_FILE_SIZE_MB:-50} depends_on: redis: condition: service_started postgres: condition: service_healthy docreader: condition: service_healthy networks: - crm-network restart: unless-stopped extra_hosts: - "host.docker.internal:host-gateway" docreader: image: swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/wechatopenai/weknora-docreader:v0.3.6 build: context: . dockerfile: docker/Dockerfile.docreader container_name: WeKnora-docreader ports: - "${DOCREADER_PORT:-50051}:50051" environment: - MINIO_ENDPOINT=minio:9000 - MINIO_PUBLIC_ENDPOINT=http://${HOST_IP}:${MINIO_PORT:-9000} - MINERU_ENDPOINT=${MINERU_ENDPOINT:-} - MAX_FILE_SIZE_MB=${MAX_FILE_SIZE_MB:-} healthcheck: test: ["CMD", "grpc_health_probe", "-addr=:50051"] interval: 30s timeout: 10s retries: 3 start_period: 60s networks: - crm-network restart: unless-stopped extra_hosts: - "host.docker.internal:host-gateway" postgres: image: paradedb/paradedb:v0.21.4-pg17 container_name: WeKnora-postgres environment: - POSTGRES_USER=${DB_USER} - POSTGRES_PASSWORD=${DB_PASSWORD} - POSTGRES_DB=${DB_NAME} ports: - "15432:5432" volumes: - postgres-data:/var/lib/postgresql/data - ./db/init.sql:/docker-entrypoint-initdb.d/init.sql networks: - crm-network healthcheck: test: ["CMD-SHELL", "pg_isready -U ${DB_USER}"] interval: 10s timeout: 10s retries: 3 start_period: 30s restart: unless-stopped stop_grace_period: 1m redis: image: redis:7.0-alpine container_name: WeKnora-redis command: redis-server --appendonly yes --requirepass ${REDIS_PASSWORD} restart: always healthcheck: test: [ "CMD-SHELL", "redis-cli -a ${REDIS_PASSWORD} ping | grep -q PONG", ] interval: 10s timeout: 30s retries: 5 start_period: 10s networks: - crm-network minio: image: swr.cn-north-4.myhuaweicloud.com/ddn-k8s/docker.io/minio/minio:RELEASE.2024-10-02T17-50-41Z container_name: WeKnora-minio ports: - "${MINIO_PORT:-9000}:9000" - "${MINIO_CONSOLE_PORT:-9001}:9001" environment: - MINIO_ROOT_USER=${MINIO_ACCESS_KEY_ID:-minioadmin} - MINIO_ROOT_PASSWORD=${MINIO_SECRET_ACCESS_KEY:-minioadmin} - MINIO_IDENTITY_OPENID_CONFIG_URL=http://${HOST_IP}/.well-known/openid-configuration - MINIO_IDENTITY_OPENID_CLIENT_ID=minio-console - MINIO_IDENTITY_OPENID_CLIENT_SECRET=minio-console-secret-key-2024 - MINIO_IDENTITY_OPENID_CLAIM_NAME=policy - MINIO_IDENTITY_OPENID_SCOPES=openid,profile,email - MINIO_IDENTITY_OPENID_REDIRECT_URI=http://${HOST_IP}:9001/oauth_callback - MINIO_IDENTITY_OPENID_DISPLAY_NAME=AI CRM Login command: server --console-address ":9001" /data volumes: - minio_data:/data healthcheck: test: ["CMD", "curl", "-f", "http://localhost:9000/minio/health/live"] interval: 30s timeout: 20s retries: 3 depends_on: crm: condition: service_healthy frontend: condition: service_started networks: - crm-network crm: build: context: .. dockerfile: Dockerfile.crm image: wk_ai_crm:local container_name: crm platform: linux/amd64 ports: - "8088:8088" restart: always volumes: - ./aicrm/logs:/opt/wk_ai_crm/logs - ./aicrm/config:/opt/wk_ai_crm/config depends_on: redis: condition: service_healthy postgres: condition: service_healthy environment: - LANG=zh_CN.UTF-8 - LANGUAGE=zh_CN:zh - LC_ALL=zh_CN.UTF-8 - HOST_IP=${HOST_IP} - BASE_PATH=${CRM_BASE_PATH:-/} - MAIL_CREDENTIAL_ENCRYPTION_KEY=${MAIL_CREDENTIAL_ENCRYPTION_KEY:-aicrm-single-mail-key-change-me} - MAIL_PROXY_ENABLED=${MAIL_PROXY_ENABLED:-false} - MAIL_PROXY_URL=${MAIL_PROXY_URL:-} healthcheck: test: ["CMD", "curl", "-f", "http://localhost:8088/"] interval: 30s timeout: 60s retries: 3 networks: - crm-network networks: crm-network: driver: bridge volumes: postgres-data: data-files: minio_data:五、环境变量配置 (.env)在 docker/ 目录下创建 .env 文件：# ========== 基础配置 ========== GIN_MODE=debug # debug(开发) / release(生产) DISABLE_REGISTRATION=false # 生产环境建议设为 true # Ollama（可选，本地大模型） OLLAMA_BASE_URL=http://host.docker.internal:11434 # ========== 存储配置 ========== DB_DRIVER=postgres RETRIEVE_DRIVER=postgres # 向量存储：postgres / elasticsearch_v7 / qdrant STORAGE_TYPE=local # 文件存储：local / minio / cos STREAM_MANAGER_TYPE=redis # ========== 端口配置 ========== APP_PORT=8080 FRONTEND_PORT=8082 DOCREADER_PORT=50051 # ========== 数据库 ========== DB_USER=postgres DB_PASSWORD=postgres123!@# DB_NAME=WeKnora # ========== Redis ========== REDIS_PASSWORD=redis123!@# REDIS_DB=0 REDIS_PREFIX=stream: # ========== 存储路径 ========== LOCAL_STORAGE_BASE_DIR=/data/files AUTO_RECOVER_DIRTY=true TENANT_AES_KEY=weknorarag-api-key-secret-secret # ========== 功能开关 ========== ENABLE_GRAPH_RAG=false # 知识图谱（构建耗时较长，需调用大模型） # ========== 安全配置 ========== JWT_SECRET=weknora-jwt-secret CONCURRENCY_POOL_SIZE=5 # Embedding 并发（429 错误时调小） HOST_IP=127.0.0.1 CRM_BASE_PATH=/六、与宝塔面板共存（Nginx 反向代理配置）若服务器已部署宝塔，将以下配置加入站点 Nginx 配置即可。server { listen 80; server_name abc.dxw123.top abc.dacaishen.eu.cc; index index.html index.htm; root /www/wwwroot/abc.dxw123.top; #CERT-APPLY-CHECK--START include /www/server/panel/vhost/nginx/well-known/abc.dxw123.top.conf; #CERT-APPLY-CHECK--END include /www/server/panel/vhost/nginx/extension/abc.dxw123.top/*.conf; error_page 404 /404.html; #REWRITE-START include /www/server/panel/vhost/rewrite/abc.dxw123.top.conf; #REWRITE-END # ========== 悟空 AI CRM 反向代理 START ========== # AI CRM 后端 API location ^~ /api/ { proxy_pass http://127.0.0.1:8088; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; proxy_set_header X-Forwarded-Proto $scheme; proxy_buffering off; proxy_cache off; proxy_read_timeout 300s; proxy_send_timeout 300s; client_max_body_size 100m; } # WeKnora App API location ^~ /weknora/ { proxy_pass http://127.0.0.1:8080; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; proxy_set_header X-Forwarded-Proto $scheme; proxy_buffering off; proxy_cache off; proxy_read_timeout 300s; client_max_body_size 100m; } # 全局上传限制（配合 MAX_FILE_SIZE_MB=50） client_max_body_size 60m; # AI CRM 前端 location / { proxy_pass http://127.0.0.1:8082; proxy_set_header Host $host; proxy_set_header X-Real-IP $remote_addr; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; proxy_set_header X-Forwarded-Proto $scheme; } # ========== 悟空 AI CRM 反向代理 END ========== # 禁止访问敏感文件 location ~* (\.user.ini|\.htaccess|\.htpasswd|\.env.*|\.gitignore|\.DS_Store|\.bashrc|README\.md|Dockerfile|docker-compose\.yml|\.sql(\.gz)?)$ { return 404; } # 禁止访问敏感目录 location ~* /(\.git|\.svn|\.vscode|\.idea|\.ssh|\.github|node_modules|runtime)/ { return 404; } # SSL 证书验证 location ~ \.well-known { allow all; } if ( $uri ~ "^/\.well-known/.*\.(php|jsp|py|js|css|lua|ts|go|zip|tar\.gz|rar|7z|sql|bak)$" ) { return 403; } access_log /www/wwwlogs/abc.dxw123.top.log; error_log /www/wwwlogs/abc.dxw123.top.error.log; }代理路径说明路径代理目标说明/api/127.0.0.1:8088AI CRM 后端 API/weknora/127.0.0.1:8080WeKnora App API/127.0.0.1:8082AI CRM 前端七、登录用户名密码adminadmin123直接访问绑定的域名或 IP 即可登录管理系统。

2026 最新实战：在单卡 48GB GPU 上部署 Qwen3.5-35B-A3B MoE 模型（vLLM + Open WebUI 完整指南） 一、为什么选择 Qwen3.5-35B-A3B？Qwen3.5-35B-A3B 是阿里通义千问团队于 2026 年初发布的 混合专家（MoE）模型，具备以下优势：小体积，大能力：总参数量 35B，但每次推理仅激活约 3B 参数，显存占用远低于同级别 Dense 模型。超长上下文：原生支持 262,144 tokens，轻松处理长文档、代码库。开源免费：采用 Apache 2.0 协议，可商用，权重已在 ModelScope 魔搭社区 公开。性能卓越：在多项基准测试中超越前代 Qwen3-235B-A22B，推理成本更低。对于拥有 单张 48GB GPU（如 A6000、RTX 6000 Ada）的用户，它是目前能本地部署的 最强 MoE 模型。二、环境配置（无需虚拟环境，全局安装即可）重要提示：经大量用户反馈，vLLM 与 Open WebUI 必须在同一 Python 环境中运行，否则会出现模块缺失或 API 不兼容问题。因此，直接使用全局环境是最简单可靠的方案。# 升级 pip（避免旧版 pip 的编译问题） pip install --upgrade pip #安装魔塔社区的包 pip install modelscope # 安装 vLLM（自动匹配 CUDA 版本） pip install vllm # 安装 FlashAttention（加速注意力计算） pip install flash-attn --no-build-isolation # 安装webui聊天界面 pip install open-webui三、单卡部署：Qwen3.5-35B-A3B 快速启动1. 下载模型（以 Qwen3.5-35B-A3B 为例）#模型下载 from modelscope import snapshot_download model_dir = snapshot_download('Qwen3.5-35B-A3B',cache_dir='./')四、单卡部署：启动 vLLM 服务vllm serve ./Qwen3.5-35B-A3B \ --dtype bfloat16 \ --port 5000 \ --max_model_len 262144 \ --gpu_memory_utilization 0.85参数详解（针对 MoE 模型优化）：参数推荐值说明--dtypebfloat16MoE 模型必须用 bfloat16，auto 可能导致精度下降--port5000API 服务端口（可自定义）--max_model_len262144启用模型最大上下文长度（爆显存时可降至 131072）--gpu_memory_utilization0.8548GB 显存安全值（若爆显存，逐步降至 0.8 → 0.75）显存占用参考（48GB GPU）：上下文 262K：~41GB上下文 131K：~38GB上下文 65K：~35GB五、集成 Open WebUI（图形化聊天界面）# 启用网络加速（AutoDL 用户必备） source /etc/network_turbo # 设置环境变量（关键！） export HF_ENDPOINT=https://hf-mirror.com # Hugging Face 镜像加速 export ENABLE_OLLAMA_API=False # 禁用 Ollama 兼容层 export OPENAI_API_BASE_URL=http://127.0.0.1:5000/v1 # 指向 vLLM API export DEFAULT_MODELS="Qwen3.5-35B-A3B" # 必须与模型文件夹名一致 # 启动 WebUI（默认端口 8080，此处改为 6006） open-webui serve --port 6006访问界面浏览器打开：👉 http://你的服务器IP:6006六、多卡部署（扩展参考）注意：Qwen3.5-35B-A3B 仅支持张量并行（Tensor Parallelism），不支持流水线并行。# 2 卡示例（如 2×A6000） vllm serve ./Qwen3.5-35B-A3B \ --dtype bfloat16 \ --port 5000 \ --tensor-parallel-size 2 \ --gpu_memory_utilization 0.8 \ --max_model_len 262144并行策略：--tensor-parallel-size N：必须等于 GPU 数量（如 2 卡设为 2）。不要设置 --pipeline-parallel-size：MoE 模型不兼容。七、OpenAI 格式 API 调用启动兼容 API 服务（等效于 vllm serve）python -m vllm.entrypoints.openai.api_server \ --served-model-name Qwen3.5-35B-A3B \ --model ./Qwen3.5-35B-A3B \ --dtype bfloat16 \ --port 5000 \ --max_model_len 262144 \ --gpu_memory_utilization 0.85Python 调用示例from openai import OpenAI client = OpenAI( base_url="http://localhost:5000/v1", api_key="not-needed" ) response = client.chat.completions.create( model="Qwen3.5-35B-A3B", messages=[{"role": "user", "content": "请用 262K 上下文分析以下代码库..."}], max_tokens=500 ) print(response.choices[0].message.content)