1.前言

Coze Studio 开源版是字节跳动推出的Coze平台的一个开源版本，旨在为开发者和用户提供一个低代码、无代码的AI应用开发平台。Coze Studio 开源版的核心目标是简化AI机器人的开发流程，使用户能够快速构建基于AI模型的问答Bot、智能体、AI应用和插件，并将其部署到各种社交平台和即时通讯应用中。

2025年7月26日凌晨13:40左右群里收到一条消息：字节跳动的扣子（coze）正式开源了。

开源地址：https://github.com/coze-dev

其中两个比较重要的仓库：

• Coze Studio 一站式 AI Agent 开发工具
• https://github.com/coze-dev/coze-studio
• Coze Loop 一个面向开发者，专注于 AI Agent 开发与运维的平台级解决方案
• https://github.com/coze-dev/cozeloop

此次扣子的开源动作堪称“全盘托出”。两大核心组件均以Apache 2.0协议开放，允许免费商用，打破过往封闭生态的商业模。

既然开源了我们带大家先本地环境安装体验一下。话不多说，我们直接上手开干。

2.本地安装coze

源码下载

首选我们需要到下载源码

# 克隆代码 git clone https://github.com/coze-dev/coze-studio.git

源码下载完成我们先看一下代码结构。这个代码前端是TypeScript 写的，后端是Go 写的。这个效率方面应该是高于dify (直观的感受)

代码这里就带大家解读了，大家也不关心这个。

模型配置

在部署之前我们需要用到模型，从配置文件里可以看到它支持OpenAI（含 Azure）、火山方舟（Ark）、DeepSeek、Ollama（本地开源模型）、Qwen（通义千问）、Gemini（Google）等几个主流的模型。

我这里就以字节提供的doubao-seed-1.6模型介绍。我先到获取火山方舟 API 获取apikey

在开通管理这里，开通一些模型（目前字节在推广用送一些模型token给大家体验）

我们需要开通需要用到的模型。开通模型后需要用到获取 Endpoint ID，大家可以看获取 Endpoint ID，这个是字节最反人类的设计，很多小伙伴不知道这个Endpoint ID是什么，每个人还不一样。需要自己创建自定义推荐接入点

上面的ep-20250621103906-7vrrn 这个才是我们要的。记住这个值，记住这个值，记住这个值。 我们说三遍。我自己做了个对照表方便后面查找。（下面的是我自己的，你的不一样，别抄）

docker 部署

项目提供的docker 部署，我们直接到docker 目录下。

我们打开docker-compose.yml 查看(详细内容就不贴了)。主要用到了elasticsearch、etcd、minio、mysql、rocketmq 、Redis 等主流中间件。

数据持久化通过volumes 挂载卷的形式存放。熟悉dify小伙伴应该比较熟悉，内容都差不多。

通过 配置本地目录与容器目录映射，实现数据持久化：

• MySQL 数据：./data/mysql
• Redis 数据：./data/bitnami/redis
• Elasticsearch 数据：./data/bitnami/elasticsearch
• MinIO 数据：./data/minio
• Etcd 数据：./data/bitnami/etcd
• Milvus 数据：./data/milvus

服务这里用到了opencoze/opencoze:latest 比dify 服务要少一点。

配置模型。

从模板目录复制 doubao-seed-1.6 模型的模版文件，并粘贴到配置文件目录。

cd coze-studio
# 复制模型配置模版
cp backend/conf/model/template/model_template_ark_doubao-seed-1.6.yaml backend/conf/model/ark_doubao-seed-1.6.yaml

在配置文件目录下，修改模版文件。

设置 、、 字段，并保存文件。

• id：Coze Studio 中的模型 ID，由开发者自行定义，必须是非 0 的整数，且全局唯一。模型上线后请勿修改模型 id 。
• meta.conn_config.api_key：模型服务的 API Key
• meta.conn_config.model：模型服务的 model ID

上面配置完成后，我们就可以启动docker了。我们这里用windows上安装的WSL 来实现。

关于WSL 安装可以看我之前的文章《网页链接》

目前它这个开源的代码和脚本不支持windows Docker Desktop 安装，结果改了一个地方还有其他地方，所以就没继续折腾了。

一键启动我之前的WSL toggle-wsl.bat

用linux 命令行切换到上面下载的代码目录下

# 启动服务
cd docker
cp .env.example .env
#Linux 适用
docker compose --profile '*' up -d

这个启动时间会比较长，需要提前下载镜像然后在启动程序。国内小伙伴可能会遇到下载镜像下载不了的问题。大家也可以借助容器代理来间接来解决问题，这里就不详细展开了。关于镜像加速可以参考这个文章 https://blog.xuanyuan.me/archives/1154

在启动docker 容器会遇到这下面这些问题

具体详细问题可以参考github上 https://github.com/coze-dev/coze-studio/issues/28

这里我们就不做详细展开。我们可以使用我改造后的docker-compose.yml 来解决上述的问题

另外也要注意和本地电脑上某些应用端口比要冲突，比如我的3306端口已经被数据库使用了。我改成5306

minio 端口也是 我修改成9901

大家可以根据自己电脑实际情况修改。

可以用我这个docker-compose.yml 来覆盖原来的

x-env-file: &env_file
 - .env
services:
 mysql:
   image: mysql:8.4.5
   container_name: coze-mysql
   environment:
     MYSQL_ROOT_PASSWORD: ${MYSQL_ROOT_PASSWORD:-root}
     MYSQL_DATABASE: ${MYSQL_DATABASE:-opencoze}
     MYSQL_USER: ${MYSQL_USER:-coze}
     MYSQL_PASSWORD: ${MYSQL_PASSWORD:-coze123}
   profiles: ['middleware', 'mysql-setup', 'mysql']
   env_file: *env_file
   ports:
     - '3306:3306'
   volumes:
     - mysql_data:/var/lib/mysql
   command:
     - --character-set-server=utf8mb4
     - --collation-server=utf8mb4_unicode_ci
   healthcheck:
     test:
       [
         'CMD',
         'mysqladmin',
         'ping',
         '-h',
         'localhost',
         '-u$${MYSQL_USER}',
         '-p$${MYSQL_PASSWORD}',
       ]
     interval: 10s
     timeout: 5s
     retries: 5
     start_period: 30s
   networks:
     - coze-network
 redis:
   image: bitnami/redis:8.0
   container_name: coze-redis
   user: root
   privileged: true
   profiles: ['middleware']
   env_file: *env_file
   environment:
     - REDIS_AOF_ENABLED=${REDIS_AOF_ENABLED:-no}
     - REDIS_PORT_NUMBER=${REDIS_PORT_NUMBER:-6379}
     - REDIS_IO_THREADS=${REDIS_IO_THREADS:-4}
     - ALLOW_EMPTY_PASSWORD=${ALLOW_EMPTY_PASSWORD:-yes}
   ports:
     - '6379:6379'
   volumes:
     - redis_data:/bitnami/redis/data:rw,Z
   command: >
     bash -c "
       /opt/bitnami/scripts/redis/setup.sh
       # Set proper permissions for data directories
       chown -R redis:redis /bitnami/redis/data
       chmod g+s /bitnami/redis/data
       exec /opt/bitnami/scripts/redis/entrypoint.sh /opt/bitnami/scripts/redis/run.sh
     "
   depends_on:
     minio-setup:
       condition: service_completed_successfully
     elasticsearch-setup:
       condition: service_completed_successfully
     mysql-setup-schema:
       condition: service_completed_successfully
     mysql-setup-init-sql:
       condition: service_completed_successfully
   healthcheck:
     test: ['CMD', 'redis-cli', 'ping']
     interval: 5s
     timeout: 10s
     retries: 10
     start_period: 10s
   networks:
     - coze-network
 elasticsearch:
   image: bitnami/elasticsearch:8.18.0
   container_name: coze-elasticsearch
   user: root
   privileged: true
   profiles: ['middleware']
   env_file: *env_file
   environment:
     - TEST=1
     # Add Java certificate trust configuration
     # - ES_JAVA_OPTS=-Djdk.tls.client.protocols=TLSv1.2 -Dhttps.protocols=TLSv1.2 -Djavax.net.ssl.trustAll=true -Xms4096m -Xmx4096m
   ports:
     - '9200:9200'
   volumes:
     - elasticsearch_data:/bitnami/elasticsearch/data
     - ./volumes/elasticsearch/elasticsearch.yml:/tmp/my_elasticsearch.yml
     - ./volumes/elasticsearch/analysis-smartcn.zip:/opt/bitnami/elasticsearch/analysis-smartcn.zip:rw,Z
   healthcheck:
     test:
       [
         'CMD-SHELL',
         'curl -f http://localhost:9200 && [ -f /tmp/es_plugins_ready ]',
       ]
     interval: 5s
     timeout: 10s
     retries: 10
     start_period: 10s
   networks:
     - coze-network
   # Install smartcn analyzer plugin
   command: >
     bash -c "
       /opt/bitnami/scripts/elasticsearch/setup.sh
       # Set proper permissions for data directories
       chown -R elasticsearch:elasticsearch /bitnami/elasticsearch/data
       chmod g+s /bitnami/elasticsearch/data
       cp /tmp/my_elasticsearch.yml /opt/bitnami/elasticsearch/config/my_elasticsearch.yml
       # Create plugin directory
       mkdir -p /bitnami/elasticsearch/plugins;
       # Unzip plugin to plugin directory and set correct permissions
       echo 'Installing smartcn plugin...';
       if [ ! -d /opt/bitnami/elasticsearch/plugins/analysis-smartcn ]; then
         # Download plugin package locally
         echo 'Copying smartcn plugin...';
         cp /opt/bitnami/elasticsearch/analysis-smartcn.zip /tmp/analysis-smartcn.zip
         elasticsearch-plugin install file:///tmp/analysis-smartcn.zip
         if [[ "$?" != "0" ]]; then
           echo 'Plugin installation failed, exiting operation';
           rm -rf /opt/bitnami/elasticsearch/plugins/analysis-smartcn
           exit 1;
         fi;
         rm -f /tmp/analysis-smartcn.zip;
       fi;
       # Create marker file indicating plugin installation success
       touch /tmp/es_plugins_ready;
       echo 'Plugin installation successful, marker file created';
       # Start Elasticsearch
       exec /opt/bitnami/scripts/elasticsearch/entrypoint.sh /opt/bitnami/scripts/elasticsearch/run.sh
       echo -e "⏳ Adjusting Elasticsearch disk watermark settings..."
     "
 minio:
   image: minio/minio:RELEASE.2025-06-13T11-33-47Z-cpuv1
   container_name: coze-minio
   user: root
   privileged: true
   profiles: ['middleware']
   env_file: *env_file
   ports:
     - '9000:9000'
     - '9001:9001'
   volumes:
     - ./data/minio:/data
   environment:
     MINIO_ROOT_USER: ${MINIO_ROOT_USER:-minioadmin}
     MINIO_ROOT_PASSWORD: ${MINIO_ROOT_PASSWORD:-minioadmin123}
     MINIO_DEFAULT_BUCKETS: ${MINIO_BUCKET:-opencoze},${MINIO_DEFAULT_BUCKETS:-milvus}
   command: server /data --console-address ":9001"
   healthcheck:
     test:
       [
         'CMD-SHELL',
         '/usr/bin/mc alias set health_check http://localhost:9000 ${MINIO_ROOT_USER} ${MINIO_ROOT_PASSWORD} && /usr/bin/mc ready health_check',
       ]
     interval: 30s
     timeout: 10s
     retries: 3
     start_period: 30s
   networks:
     - coze-network
 etcd:
   image: bitnami/etcd:3.5
   container_name: coze-etcd
   user: root
   privileged: true
   profiles: ['middleware']
   env_file: *env_file
   environment:
     - ETCD_AUTO_COMPACTION_MODE=revision
     - ETCD_AUTO_COMPACTION_RETENTION=1000
     - ETCD_QUOTA_BACKEND_BYTES=4294967296
     - ALLOW_NONE_AUTHENTICATION=yes
   ports:
     - 2379:2379
     - 2380:2380
   volumes:
     - ./data/bitnami/etcd:/bitnami/etcd:rw,Z
     - ./volumes/etcd/etcd.conf.yml:/opt/bitnami/etcd/conf/etcd.conf.yml:ro,Z
   command: >
     bash -c "
       /opt/bitnami/scripts/etcd/setup.sh
       # Set proper permissions for data and config directories
       chown -R etcd:etcd /bitnami/etcd
       chmod g+s /bitnami/etcd
       exec /opt/bitnami/scripts/etcd/entrypoint.sh /opt/bitnami/scripts/etcd/run.sh
     "
   healthcheck:
     test: ['CMD', 'etcdctl', 'endpoint', 'health']
     interval: 5s
     timeout: 10s
     retries: 10
     start_period: 10s
   networks:
     - coze-network
 milvus:
   container_name: coze-milvus
   image: milvusdb/milvus:v2.5.10
   user: root
   privileged: true
   profiles: ['middleware']
   env_file: *env_file
   command: >
     bash -c "
       # Set proper permissions for data directories
       chown -R root:root /var/lib/milvus
       chmod g+s /var/lib/milvus
       exec milvus run standalone
     "
   security_opt:
     - seccomp:unconfined
   environment:
     ETCD_ENDPOINTS: coze-etcd:2379
     MINIO_ADDRESS: coze-minio:9000
     MINIO_BUCKET_NAME: ${MINIO_BUCKET:-milvus}
     MINIO_ACCESS_KEY_ID: ${MINIO_ROOT_USER:-minioadmin}
     MINIO_SECRET_ACCESS_KEY: ${MINIO_ROOT_PASSWORD:-minioadmin123}
     MINIO_USE_SSL: false
     LOG_LEVEL: debug
   volumes:
     - ./data/milvus:/var/lib/milvus:rw,Z
   healthcheck:
     test: ['CMD', 'curl', '-f', 'http://localhost:9091/healthz']
     interval: 5s
     timeout: 10s
     retries: 10
     start_period: 10s
   ports:
     - '19530:19530'
     - '9091:9091'
   depends_on:
     etcd:
       condition: service_healthy
     minio:
       condition: service_healthy
   networks:
     - coze-network
 nsqlookupd:
   image: nsqio/nsq:v1.2.1
   container_name: coze-nsqlookupd
   command: /nsqlookupd
   profiles: ['middleware']
   ports:
     - '4160:4160'
     - '4161:4161'
   networks:
     - coze-network
   healthcheck:
     test: ['CMD-SHELL', 'nsqlookupd --version']
     interval: 5s
     timeout: 10s
     retries: 10
     start_period: 10s
 nsqd:
   image: nsqio/nsq:v1.2.1
   container_name: coze-nsqd
   command: /nsqd --lookupd-tcp-address=coze-nsqlookupd:4160 --broadcast-address=coze-nsqd
   profiles: ['middleware']
   ports:
     - '4150:4150'
     - '4151:4151'
   depends_on:
     nsqlookupd:
       condition: service_healthy
   networks:
     - coze-network
   healthcheck:
     test: ['CMD-SHELL', '/nsqd --version']
     interval: 5s
     timeout: 10s
     retries: 10
     start_period: 10s
 nsqadmin:
   image: nsqio/nsq:v1.2.1
   container_name: coze-nsqadmin
   command: /nsqadmin --lookupd-http-address=coze-nsqlookupd:4161
   profiles: ['middleware']
   ports:
     - '4171:4171'
   depends_on:
     nsqlookupd:
       condition: service_healthy
   networks:
     - coze-network
 elasticsearch-setup:
   image: alpine/curl:8.12.1
   container_name: coze-elasticsearch-setup
   profiles: ['middleware', 'volcano-setup']
   env_file: *env_file
   depends_on:
     elasticsearch:
       condition: service_healthy
   volumes:
     - ./volumes/elasticsearch/setup_es.sh:/setup_es.sh
     - ./volumes/elasticsearch/es_index_schema:/es_index_schema
   command:
     - /bin/sh
     - -c
     - |
       set -ex
       /setup_es.sh
       echo 'Elasticsearch setup complete.'
   networks:
     - coze-network
   restart: 'no'
 minio-setup:
   image: minio/mc:RELEASE.2025-05-21T01-59-54Z-cpuv1
   container_name: coze-minio-setup
   profiles: ['middleware']
   env_file: *env_file
   depends_on:
     minio:
       condition: service_healthy
   volumes:
     - ./volumes/minio/default_icon/:/default_icon
     - ./volumes/minio/official_plugin_icon/:/official_plugin_icon
   entrypoint: >
     /bin/sh -c  "
       (/usr/bin/mc alias set localminio http://coze-minio:9000 ${MINIO_ROOT_USER} ${MINIO_ROOT_PASSWORD} && \
       /usr/bin/mc mb --ignore-existing localminio/${STORAGE_BUCKET} && \
       /usr/bin/mc cp --recursive /default_icon/ localminio/${STORAGE_BUCKET}/default_icon/ && \
       /usr/bin/mc cp --recursive /official_plugin_icon/ localminio/${STORAGE_BUCKET}/official_plugin_icon/ && \
       echo 'upload files to minio complete: Files uploaded to ${STORAGE_BUCKET} bucket.') || exit 1; \
     "
   networks:
     - coze-network
   restart: 'no'
 mysql-setup-schema:
   image: arigaio/atlas:0.35.0-community-alpine
   container_name: coze-mysql-setup-schema
   profiles: ['middleware', 'mysql-setup', 'run-server']
   env_file: *env_file
   depends_on:
     mysql:
       condition: service_healthy
   volumes:
     - ./atlas/opencoze_latest_schema.hcl:/opencoze_latest_schema.hcl
   entrypoint:
     - /bin/sh
     - -c
     - |
       set -ex
       TMP_ATLAS_URL="${ATLAS_URL}"
       if [ "${MYSQL_HOST}" = "localhost" ] || [ "${MYSQL_HOST}" = "127.0.0.1" ]; then
           echo "MYSQL_HOST is localhost or 127.0.0.1, replacing with docker network address"
           TMP_ATLAS_URL="mysql://${MYSQL_USER}:${MYSQL_PASSWORD}@mysql:${MYSQL_PORT}/${MYSQL_DATABASE}?charset=utf8mb4&parseTime=True"
       fi
       echo "final atlas url: $${TMP_ATLAS_URL}"
       for i in `seq 1 60`; do
            if atlas schema apply \
                -u "$${TMP_ATLAS_URL}" \
                --to file:///opencoze_latest_schema.hcl \
                --exclude "atlas_schema_revisions,table_*" \
                --auto-approve; then
                echo "MySQL setup complete."
                exit 0
            fi
            echo "atlas schema apply failed, retrying...($$i/60)"
            sleep 1
        done
        echo "MySQL setup failed after 60 retries."
        exit 1
   networks:
     - coze-network
 mysql-setup-init-sql:
   image: mysql:8.4.5
   container_name: coze-mysql-setup-init-sql
   profiles: ['middleware', 'mysql-setup', 'run-server', 'volcano-setup']
   env_file: *env_file
   depends_on:
     mysql:
       condition: service_healthy
   command:
     - /bin/sh
     - -c
     - |
       set -ex
       for i in $(seq 1 60); do
         DB_HOST="$${MYSQL_HOST}"
         if [ "$${MYSQL_HOST}" = "localhost" ] || [ "$${MYSQL_HOST}" = "127.0.0.1" ]; then
           DB_HOST="mysql"
         fi
         if mysql -h "$${DB_HOST}" -P"$${MYSQL_PORT}" -u"$${MYSQL_USER}" -p"$${MYSQL_PASSWORD}" "$${MYSQL_DATABASE}" < /schema.sql && \
            mysql -h "$${DB_HOST}" -P"$${MYSQL_PORT}" -u"$${MYSQL_USER}" -p"$${MYSQL_PASSWORD}" "$${MYSQL_DATABASE}" < /sql_init.sql; then
           echo 'MySQL init success.'
           exit 0
         fi
         echo "Retrying to connect to mysql... ($$i/60)"
         sleep 1
       done
       echo 'Failed to init mysql db.'
       exit 1
   volumes:
     - ./volumes/mysql/sql_init.sql:/sql_init.sql
     - ./volumes/mysql/schema.sql:/schema.sql
   networks:
     - coze-network
   restart: 'no'
 coze-server:
   # build:
   #   context: ../
   #   dockerfile: backend/Dockerfile
   image: opencoze/opencoze:latest
   container_name: coze-server
   profiles: ['run-server']
   env_file: *env_file
   environment:
     LISTEN_ADDR: 0.0.0.0:8888
   networks:
     - coze-network
   ports:
     - '8888:8888'
     - '8889:8889'
   volumes:
     - .env:/app/.env
     - ../backend/conf:/app/resources/conf
     # - ../backend/static:/app/resources/static
   depends_on:
     mysql:
       condition: service_healthy
     redis:
       condition: service_healthy
     # rocketmq-namesrv:
     #   condition: service_healthy
     # rocketmq-broker:
     #   condition: service_healthy
     elasticsearch:
       condition: service_healthy
     minio:
       condition: service_healthy
     milvus:
       condition: service_healthy
     minio-setup:
       condition: service_completed_successfully
     elasticsearch-setup:
       condition: service_completed_successfully
     mysql-setup-init-sql:
       condition: service_completed_successfully
   command: ['/app/bootstrap.sh']
networks:
 coze-network:
   driver: bridge
volumes:
 mysql_data:
 elasticsearch_data:
 redis_data:

部署完成后我们就可以通过URL 地址访问了（http://localhost:8888/sign?redirect=%2Fspace）

3 本地coze体验

输入网址http://localhost:8888/

这样我们就看到本地搭建的COZE 了。

创建智能体

点击创建后填写名字就看到我们熟悉的COZE 了

我们之前配置文件写的是doubao的模型，大家也可以根据自己的需要修改和添加模型

我们可以将配置文件复制到F:\temp\coze-studio\backend\conf\model 目录下，添加新的模型

验证测试

接下来我们在刚才配置的智能体 做个简单聊天对话功能

好了一个简单的聊天对话的智能体就配置完成了。其他功能大家自行去研究。

4.总结

今天主要带大家了解并实现了 Coze Studio 开源版的本地安装与使用。此方案借助火山方舟等主流模型，为开发者和用户提供了一个低代码、无代码的 AI 应用开发平台，简化了 AI 机器人的开发流程。

该方案不仅解决了开发者在 AI 应用开发过程中面临的技术门槛问题，通过详细的源码下载、模型配置和 docker 部署步骤，让用户能够快速构建基于 AI 模型的问答 Bot、智能体、AI 应用和插件，并将其部署到各种社交平台和即时通讯应用中。还提供了一种可复用的本地开发与部署思路，其中模型配置及 docker 部署的过程详细且具有通用性，对于有类似需求的项目具有参考价值。

需要注意的是，在部署过程中可能会遇到一些问题，如镜像下载缓慢、脚本格式不兼容、端口冲突等，大家可以根据实际情况进行解决。另外，本文以 doubao - seed - 1.6 模型为例进行介绍，大家也可以根据自己的需要修改和添加其他模型。

感兴趣的小伙伴可以按照本文步骤去尝试本地安装和使用 Coze Studio 开源版，体验低代码、无代码的 AI 应用开发乐趣。今天的分享就到这里结束了，我们下一篇文章见。