NodeTool
Visual AI workflow builder combining ComfyUI's node-based flexibility with n8n's automation power. Build LLM agents, RAG pipelines, and multimodal data flows on your local machine.
Quick Start
CODEBLOCK0
Installation
Linux / macOS
Quick one-line installation:
CODEBLOCK1
With custom directory:
CODEBLOCK2
Non-interactive mode (automatic, no prompts):
Both scripts support silent installation:
CODEBLOCK3
What happens with non-interactive mode:
- - All confirmation prompts are skipped automatically
- Installation proceeds without requiring user input
- Perfect for CI/CD pipelines or automated setups
Windows
Quick one-line installation:
CODEBLOCK4
With custom directory:
CODEBLOCK5
Non-interactive mode:
CODEBLOCK6
Core Commands
Workflows
Manage and execute NodeTool workflows:
CODEBLOCK7
Run Options
Execute workflows in different modes:
CODEBLOCK8
Assets
Manage workflow assets (nodes, models, files):
CODEBLOCK9
Packages
Manage NodeTool packages (export workflows, generate docs):
CODEBLOCK10
Jobs
Manage background job executions:
CODEBLOCK11
Deployment
Deploy NodeTool to cloud platforms (RunPod, GCP, Docker):
CODEBLOCK12
Model Management
Discover and manage AI models (HuggingFace, Ollama):
CODEBLOCK13
Admin
Maintain model caches and clean up:
CODEBLOCK14
Chat & Server
Interactive chat and web interface:
CODEBLOCK15
Proxy
Start reverse proxy with HTTPS:
CODEBLOCK16
Other Commands
CODEBLOCK17
Use Cases
Workflow Execution
Run a NodeTool workflow and get structured output:
CODEBLOCK18
Package Creation
Generate documentation for a custom package:
CODEBLOCK19
Deployment
Deploy a NodeTool instance to the cloud:
CODEBLOCK20
Model Management
Check and manage cached AI models:
CODEBLOCK21
Installation
Linux / macOS
Quick one-line installation:
CODEBLOCK22
With custom directory:
CODEBLOCK23
Non-interactive mode (automatic, no prompts):
Both scripts support silent installation:
CODEBLOCK24
What happens with non-interactive mode:
- - All confirmation prompts are skipped automatically
- Installation proceeds without requiring user input
- Perfect for CI/CD pipelines or automated setups
Windows
Quick one-line installation:
CODEBLOCK25
With custom directory:
CODEBLOCK26
Non-interactive mode:
CODEBLOCK27
What Gets Installed
The installer sets up:
- - micromamba — Python package manager (conda replacement)
- NodeTool environment — Conda env at INLINECODE0
- Python packages —
nodetool-core, nodetool-base from NodeTool registry - Wrapper scripts —
nodetool CLI available from any terminal
Environment Setup
After installation, these variables are automatically configured:
CODEBLOCK28
System Info
Check NodeTool environment and installed packages:
CODEBLOCK29
Output shows:
- - Version
- Python version
- Platform/Architecture
- Installed AI packages (OpenAI, Anthropic, Google, HF, Ollama, fal-client)
- Environment variables
- API key status
NodeTool
可视化AI工作流构建器,融合了ComfyUI基于节点的灵活性与n8n的自动化能力。在本地机器上构建LLM代理、RAG流水线以及多模态数据流。
快速开始
bash
查看系统信息
nodetool info
列出工作流
nodetool workflows list
交互式运行工作流
nodetool run
启动聊天界面
nodetool chat
启动Web服务器
nodetool serve
安装
Linux / macOS
一行快速安装:
bash
curl -fsSL https://raw.githubusercontent.com/nodetool-ai/nodetool/refs/heads/main/install.sh | bash
使用自定义目录:
bash
curl -fsSL https://raw.githubusercontent.com/nodetool-ai/nodetool/refs/heads/main/install.sh | bash --prefix ~/.nodetool
非交互模式(自动,无提示):
两个脚本均支持静默安装:
bash
Linux/macOS - 使用 -y
curl -fsSL https://raw.githubusercontent.com/nodetool-ai/nodetool/refs/heads/main/install.sh | bash -y
Windows - 使用 -Yes
irm https://raw.githubusercontent.com/nodetool-ai/nodetool/refs/heads/main/install.ps1 | iex; .\install.ps1 -Yes
非交互模式的作用:
- - 自动跳过所有确认提示
- 无需用户输入即可继续安装
- 完美适用于CI/CD流水线或自动化设置
Windows
一行快速安装:
powershell
irm https://raw.githubusercontent.com/nodetool-ai/nodetool/refs/heads/main/install.ps1 | iex
使用自定义目录:
powershell
.\install.ps1 -Prefix C:\nodetool
非交互模式:
powershell
.\install.ps1 -Yes
核心命令
工作流
管理和执行NodeTool工作流:
bash
列出所有工作流(用户 + 示例)
nodetool workflows list
获取特定工作流的详细信息
nodetool workflows get
按ID运行工作流
nodetool run
从文件运行工作流
nodetool run workflow.json
以JSONL格式输出运行(用于自动化)
nodetool run --jsonl
运行选项
以不同模式执行工作流:
bash
交互模式(默认)- 美观输出
nodetool run workflow_abc123
JSONL模式 - 流式JSON,供子进程使用
nodetool run workflow_abc123 --jsonl
Stdin模式 - 通过管道传入RunJobRequest JSON
echo {workflowid:abc,userid:1,auth_token:token,params:{}} | nodetool run --stdin --jsonl
使用自定义用户ID
nodetool run workflowabc123 --user-id customuser_id
使用认证令牌
nodetool run workflowabc123 --auth-token myauth_token
资源
管理工作流资源(节点、模型、文件):
bash
列出所有资源
nodetool assets list
获取资源详细信息
nodetool assets get
包
管理NodeTool包(导出工作流、生成文档):
bash
列出包
nodetool package list
生成文档
nodetool package docs
生成节点文档
nodetool package node-docs
生成工作流文档(Jekyll)
nodetool package workflow-docs
扫描目录中的节点并创建包
nodetool package scan
初始化新的包项目
nodetool package init
作业
管理后台作业执行:
bash
列出用户的作业
nodetool jobs list
获取作业详细信息
nodetool jobs get
获取作业日志
nodetool jobs logs
启动工作流的后台作业
nodetool jobs start
部署
将NodeTool部署到云平台(RunPod、GCP、Docker):
bash
初始化deployment.yaml
nodetool deploy init
列出部署
nodetool deploy list
添加新部署
nodetool deploy add
应用部署配置
nodetool deploy apply
检查部署状态
nodetool deploy status
查看部署日志
nodetool deploy logs
销毁部署
nodetool deploy destroy
管理已部署实例上的集合
nodetool deploy collections
管理已部署实例上的数据库
nodetool deploy database
管理已部署实例上的工作流
nodetool deploy workflows
查看将要进行的更改
nodetool deploy plan
模型管理
发现和管理AI模型(HuggingFace、Ollama):
bash
按类型列出缓存的HuggingFace模型
nodetool model list-hf
列出所有HuggingFace缓存条目
nodetool model list-hf-all
列出支持的HF类型
nodetool model hf-types
检查HuggingFace缓存
nodetool model hf-cache
扫描缓存获取信息
nodetool admin scan-cache
管理
维护模型缓存并进行清理:
bash
计算总缓存大小
nodetool admin cache-size
从缓存中删除HuggingFace模型
nodetool admin delete-hf
下载HuggingFace模型并显示进度
nodetool admin download-hf
下载Ollama模型
nodetool admin download-ollama
聊天与服务器
交互式聊天和Web界面:
bash
启动CLI聊天
nodetool chat
启动聊天服务器(WebSocket + SSE)
nodetool chat-server
启动FastAPI后端服务器
nodetool serve --host 0.0.0.0 --port 8000
使用静态资源文件夹
nodetool serve --static-folder ./static --apps-folder ./apps
开发模式,自动重载
nodetool serve --reload
生产模式
nodetool serve --production
代理
启动带HTTPS的反向代理:
bash
启动代理服务器
nodetool proxy
检查代理状态
nodetool proxy-status
验证代理配置
nodetool proxy-validate-config
运行代理守护进程,支持ACME HTTP + HTTPS
nodetool proxy-daemon
其他命令
bash
查看设置和密钥
nodetool settings show
为工作流生成自定义HTML应用
nodetool vibecoding
运行工作流并导出为Python DSL
nodetool dsl-export
将工作流导出为Gradio应用
nodetool gradio-export
重新生成DSL
nodetool codegen
管理数据库迁移
nodetool migrations
与远程同步数据库
nodetool sync
使用场景
工作流执行
运行NodeTool工作流并获取结构化输出:
bash
交互式运行工作流
nodetool run myworkflowid
运行并流式输出JSONL
nodetool run myworkflowid --jsonl | jq -r .[] | \(.status) | \(.output)
包创建
为自定义包生成文档:
bash
扫描节点并创建包
nodetool package scan
生成完整文档
nodetool package docs
部署
将NodeTool实例部署到云端:
bash
初始化部署配置
nodetool deploy init
添加RunPod部署
nodetool deploy add
部署并启动
nodetool deploy apply
模型管理
检查和管理缓存的AI模型:
bash
列出所有可用模型
nodetool model list-hf-all
检查缓存
nodetool model hf-cache
安装
Linux / macOS
一行快速安装:
bash
curl -fsSL https://raw.githubusercontent.com/nodetool-ai/nodetool/refs/heads/main/install.sh | bash
使用自定义目录:
bash
curl -fsSL https://raw.githubusercontent.com/nodetool-ai/nodetool/refs/heads/main/install.sh | bash --prefix ~/.nodetool
非交互模式(自动,无提示):
两个脚本均支持静默安装:
bash
Linux/macOS - 使用 -y
curl -fsSL https://raw.githubusercontent.com/nodetool-ai/nodetool/ref