AGPL-3.0 开源多模态 GenAI 工作流工作室（画布 + Python 插件引擎），以 dsh-tongflow 工作室插件接入 DeepSeek Harness，支持桌面应用、Docker 或源码三种运行方式。
DSH 适配
兼容
自动推断
安全审计
未审计
最后核验
2026-08-21
许可证
AGPL-3.0
01它能帮你完成什么？
Design multimodal generative workflows that turn text into images, talking-head videos and music videos
Finished media assets from one canvas — fused images, a lip-synced talking-head avatar video, or a complete music video (lyrics + song + characters + scenes + storyboard)
Creators and content teams who want generative-AI workflow automation across text, image, video and audio without writing code
Automate end-to-end pipeline execution instead of re-running steps manually
Workflows built once from AI-model nodes run node by node, or in one click via Execute Mode's run button
Operators and teams who need repeatable, one-click workflow automation over multimodal generation tasks
Self-host a fully local, account-free workflow automation studio
A running studio at http://localhost:3000 with plugins and credentials managed in-app; all data kept in a local /data volume (SQLite db, uploads, settings)
Individuals and teams who want their own workflow-automation studio without cloud accounts or external data storage
02如何接入 DeepSeek Harness？
前置条件
- Desktop path: macOS (Universal — Apple Silicon & Intel) or Windows x64 machine, plus a Google or WeChat account to sign in
- Self-host from source: Node with pnpm, and a Python 3.10+ interpreter on PATH
- Self-host with Docker: a Docker runtime — no Node / Python / pnpm setup required
安装步骤
- 01
Download the installer for your platform (TongFlow-mac-universal.dmg or TongFlow-win-x64.msi from GitHub Releases), install it, and open it
- 02
Sign in with Google or WeChat and start creating — the cloud studio manages plugins and execution
- 03
Self-host from source: run `pnpm install`, `pnpm plugins:install`, then `pnpm start:prod` and open http://localhost:3000
- 04
Self-host with Docker: `docker run -d -p 3000:3000 -v tongflow-data:/data -v tongflow-plugins:/plugins ghcr.io/tong-io/tongflow:latest`, then open http://localhost:3000
- 05
Self-host: install the capability plugins you need from the in-app plugin manager, and add credentials in Settings
验证接入成功
- Open http://localhost:3000 and the canvas is live (source or Docker self-host)
03DSH 适配与能力边界
dsh-tongflow — the DeepSeek Harness studio plugin listed in the repo description and topics; TongFlow's plugin-based architecture lets every platform package its own independent plugin around the studio
Multimodal canvas workflows (add / transform / combine)
text, images, audio, video, documents, URLs and 3D models added as canvas nodes→chained generative results — images, video, music, speech, 3D — arranged with just three operations, no manual node connecting
generation nodes call external model providers through the installed pluginsBroad modality coverage (text, image, video, audio, 3D)
any supported modality or file format→generation, editing, understanding, upscaling, lip sync, motion capture, music creation, OCR, speech synthesis/recognition and more as out-of-the-box nodes
Official plugin catalog — API, router and GPU/CPU plugins
plugins for OpenAI, Gemini, DeepSeek, ByteDance, xAI, Runway, OpenRouter, Replicate, fal and Modal-based GPU/CPU nodes→per-node model pickers and capability implementations installed from the in-app plugin manager, usable immediately without restart
first plugin install needs network access to GitHubofficial GPU/CPU plugins run on Modal and need MODAL_TOKEN_ID / MODAL_TOKEN_SECRETCustom plugins via ABI contract + Python SDK
a small Python package annotated against the ABI-generated types via the tongflow Python SDK→new capability nodes that pick one or more ABI slots and supply the implementation
on first run a plugin provisions a shared Python venv and installs its requirements.txt from PyPI (needs network)
04适合谁？何时不该用？
适合
- Creators and content teams who want generative-AI workflow automation across text, image, video and audio without writing code
- Operators and teams who need repeatable, one-click workflow automation over multimodal generation tasks
- Individuals and teams who want their own workflow-automation studio without cloud accounts or external data storage
不适合
- macOS builds are not yet notarized with Apple, so Gatekeeper blocks the first launch; you must clear the quarantine flag once with `xattr -cr /Applications/TongFlow.app`.
- The self-host image ships no plugins — the first plugin install needs network access to GitHub, and the first run installs Python dependencies from PyPI; Modal-backed plugins additionally need a Modal token.
05兼容性、维护与安全提示
- macOS builds are not yet notarized with Apple, so Gatekeeper blocks the first launch; you must clear the quarantine flag once with `xattr -cr /Applications/TongFlow.app`.
- The self-host image ships no plugins — the first plugin install needs network access to GitHub, and the first run installs Python dependencies from PyPI; Modal-backed plugins additionally need a Modal token.
- Capability nodes execute through external model providers (OpenAI / Gemini / OpenRouter-style API plugins, or Modal GPU), so real workloads need provider API keys or a Modal account — Modal's free tier is capped at $30/month of GPU compute.
AGPL-3.0 · actively maintained (latest release v0.3.4, 2026-08-19)
06常见问题
TongFlow 如何接入 DeepSeek Harness？
仓库提供 dsh-tongflow，即 DeepSeek Harness 工作室插件（见仓库描述与话题标签）。TongFlow 的插件化设计允许每个平台打包自己独立的插件，本页基于 TongFlow 工作室的 README 撰写，涵盖桌面应用、自托管与插件体系。
TongFlow 怎么安装？
桌面端：从 GitHub Releases 下载 macOS 通用版 dmg 或 Windows msi，安装后打开，用 Google 或微信登录即可。自托管：执行 `docker run -d -p 3000:3000 -v tongflow-data:/data -v tongflow-plugins:/plugins ghcr.io/tong-io/tongflow:latest`；或从源码运行 `pnpm install`、`pnpm plugins:install`、`pnpm start:prod`。
需要 API Key 吗？
Key 可选且按插件配置：在应用内 Settings 对话框或启动参数里填写，支持 OPENROUTER_API_KEY、GEMINI_API_KEY、OPENAI_API_KEY、MODAL_TOKEN_ID / MODAL_TOKEN_SECRET。官方 GPU/CPU 插件运行在 Modal 上，每月有最高 $30 的免费 GPU 算力额度。
macOS 首次打开被拦截怎么办？
构建尚未完成 Apple 公证，Gatekeeper 会提示“TongFlow 已损坏”。把应用移入「应用程序」后执行一次 `xattr -cr /Applications/TongFlow.app` 清除隔离属性即可正常打开，安装包请直接从 Releases 页面下载。
自托管时数据存在哪里？
所有可写内容都在 /data 卷里 —— SQLite 数据库、上传文件和设置。插件凭据由应用内 Settings 编辑器保存在本地，修改后立即生效无需重启；自托管即可获得完全本地、无需账号的 TongFlow。
07相关的 DSH 工作流
archify
作者 tt-a1i
为编码智能体生成美观可验证的架构图、时序图与数据流图，输出自包含 HTML，支持动效与清晰导出。
openviking
作者 volcengine
为 AI 智能体打造的自进化上下文数据库，统一智能体记忆、知识 RAG 与技能。
nocobase
作者 nocobase
开源的 AI + 无代码应用搭建平台：AI 在久经生产验证的底层基础设施与所见即所得的无代码界面之上工作，帮你快速构建 CRM、ERP 等业务系统，兼顾速度与可靠性。
learn-harness-engineering
作者 walkinglabs
Harness 工程新手教程，从 0 到 1 系统学习智能体工作流框架。
08数据与来源
TongFlow’s plugin-based design lets every platform package its own independent plugins, and we provide at least one offi…
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-08-21。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。