Skip to main content

tongflow/packages/dsh-tongflow

887Stars118Forks7Issues91Watchers

Embed TongFlow media workflow editor in DSH: each asset is generated by an independent workflow file, with the AI Agent handling project structure and asset management.

Evidence5/5methodologySourceInstallMaintenanceDSH versionSecurity scan
Machine-auditedInstall commandRepo verifieddsh-plugin topicLicenseREADMEAI wiki
Language
TypeScript
License
AGPL-3.0
Branch
main
3dagentaiai-toolsaigccanvasdeepseek-harnessdocument

Install

cmdweb profile
$ dsh plugin --profile web add dsh-tongflow

Run the command above in your terminal to install this plugin via the dsh CLI. You can switch Profile in the top-right corner. New to dsh? Read the beginner tutorial

Install via your agent

Install the DeepSeek Harness plugin tong-io/tongflow/packages/dsh-tongflow for me: review the repository at https://github.com/tong-io/tongflow first, then run the install command and verify the plugin loads successfully.

Paste this instruction to the DSH Web GUI assistant — it will install and verify for you.

One-line Description

Embed the TongFlow media workflow editor into DSH: when users create images, audio, video, or 3D content using the DSH Agent, each asset is generated from an independent workflow file, with the Agent responsible for project structure and asset organization. Users can directly fine-tune on the canvas and rerun.

Core Capabilities

  • Create Studio project folders with custom structure (no fixed template, designed collaboratively by Agent and user)
  • Drive all image, audio, video, and 3D asset generation using .tongflow.json workflow files; output files are placed alongside the workflow and do not overwrite historical results
  • Provide node directory and node descriptions so the Agent knows what node types, connection rules, configuration fields, and available plugins exist
  • Let the Agent review generated results via tongflow_look (images) and tongflow_perceive (video/audio/image understanding) before proceeding to the next step
  • Built-in paid plugin confirmation mechanism and project/plugin management tools to avoid accidental charges
  • Auto-clone official plugins at startup and maintain venv, keeping the canvas node directory in sync with the TongFlow official version

Technical Implementation

  • Language: TypeScript (node-side + browser-side CJS bundle in one)
  • Key Dependencies: @deepseek-ai/cordis, @deepseek-ai/dsh-* suite (agent / tools / skills / webserver / llm / jobs / system-prompt), tongflow (same repo workspace package), @deepseek-ai/schemastery
  • Architecture Pattern: Cordis plugin, inserting tongflow nodes into cordis via dsh.bundle.patch; use agent/pre-step event to determine if the current session enters Studio mode (first message starts with @tongflow), and only register tongflow_* tools, skills, and system prompt segments for Studio sessions
  • Entry Files: src/index.ts (Node-side apply) + src/client/index.ts (browser-side bundle), injected into host via cordis.patch.yml

Use Cases

When using DSH for storyboard layouts, short videos, ad assets, music MVs, character profile sets, 3D resource packs, or other batched, iterative AI-generated content, this plugin transforms "asset production" from one-time chat results into saveable, rerunnable, manually adjustable workflows; regular conversational needs (coding, information lookup) don't need it.

Prerequisites and Compatibility

DependencyMinimum VersionDescription
DSH>= 0.1.0-rc.7All @deepseek-ai/dsh-* companion dependencies require this version
Node>= 22.19.0Consistent with DSH's own requirements (stated in README)
Python>= 3.10First startup will automatically create ~/.dsh/tongflow/venv and install tongflow SDK; pythonPath can be specified
GitRequiredOfficial plugins are shallow-cloned at startup via git
FFmpegRecommendedUsed for video contact sheet generation

Installation

dsh plugin --profile web add github:tong-io/tongflow/packages/dsh-tongflow

Configuration Options

ConfigTypeDescriptionDefault
studioRootstringStudio data root directory, storing projects, virtual environments, cloned plugins, cache<DSH_HOME>/tongflow (i.e., ~/.dsh/tongflow)
pythonPathstringPython executable path (≥ 3.10) for creating Studio venv; leave empty to auto-detect in PATHAuto-detect (python3.13 → python3.10 → python)
sdkSpecstringpip spec to install in Studio venv, e.g., tongflow==0.3.0, or -e /path/to/sdk for developmenttongflow==0.3.0
pluginOrgstringGit organization address for official plugins (clone source)https://github.com/tong-io
pluginGitUrlsobjectCustom plugin id → git URL mapping, for community or private plugins{}
envobjectEnvironment variables injected into plugin subprocess (like API Key), recommended to use the "Plugins & Secrets" panel in Studio first{}
maxConcurrentRunsnumberMaximum concurrent Workflow runs2
httpPrefixstringHTTP route prefix used by Studio and canvas/tongflow
localestringInterface language for embedded canvasen
autoInstallOfficialbooleanWhether to auto shallow-clone missing official plugins at startup (only clone, deploy/charge on-demand)true

FAQ

Q: What's the difference between this plugin and using TongFlow SaaS directly?

A: They share the same TongFlow engine, node directory, and ABI, but this runs in your local DSH: assets land on your hard drive, with the DSH Agent leading project structure and asset planning, and support for local tools/skills collaboration; the SaaS version doesn't have the Agent layer.

Q: I installed it but don't see tongflow_ tools in my session?*

A: Studio mode is not activated by default. Tools and Studio UI are only injected when the first message in a session starts with @tongflow; other sessions remain regular DSH as before the plugin was enabled.

Q: If I accidentally run a paid plugin, will I be charged?

A: When tongflow_workflow_run detects the workflow uses paid plugins (API-paid or Modal GPU), it must receive explicit user_confirmed: true to actually execute; otherwise it returns needs_confirmation listing involved plugins, billing method, whether keys are configured, and available alternatives; local plugins don't require confirmation.

Q: Can I run an image multiple times?

A: Yes. Each run generates a new numbered output file (e.g., mei_ref.01.png, mei_ref.02.png), never overwriting historical results; run records are written to <name>.runs.json; just run again after adjusting the workflow.

Q: Can workflows reference other assets?

A: Yes. Data nodes can directly write file paths (relative to workflow file ../style.md or project root characters/mei/x.png), or use {{path}} placeholder to embed text file content into prompt at runtime; URLs are passed to plugins as-is.

Q: What modalities are supported?

A: The node directory covers image, audio, video, and 3D modalities, with respective modality/ data nodes and transfer/compose/decompose/batch executable node categories; specific available plugins depend on your installed/auto-cloned TongFlow plugin list.

Q: What if I don't have Python installed?

A: The first startup will error out if Python ≥ 3.10 is not found; you can install official Python, use uv python install, or specify pythonPath in config.

Learning Curve

Intermediate — Users need to understand the "workflow as asset" design paradigm, node directory and connection rules, and have Python, Git, and (optional) FFmpeg ready; the plugin's installation command itself is simple.

Known Issues and Limitations

  • If the system doesn't have Python ≥ 3.10 at startup, it will throw an error requiring installation or setting pythonPath (src/engine/bootstrap.ts:83-86)
  • Official plugin cloning depends on Git; when autoInstallOfficial=true, it pulls in the background asynchronously, during which the canvas node directory may be incomplete (src/studio.ts:65-72)
  • tongflow_workflow_run will refuse to execute paid plugins without user_confirmed=true, requiring user confirmation again in the current session each time (src/tools/run-tools.ts:78)
  • Plugin uninstall won't clean up projects, venv, cloned plugins, and env.json under ~/.dsh/tongflow; manual deletion required (no auto-cleanup logic found in source)

Read the usage guide →

Install steps, key points, FAQ and compatibility for this plugin — auto-derived from indexed fields.

Listing badge

Listed on deepseek-plugin.org
[![Listed on deepseek-plugin.org](https://img.shields.io/badge/listed_on-deepseek--plugin.org-007EC6)](https://deepseek-plugin.org/plugins/tong-io/tongflow/packages/dsh-tongflow)

Paste this markdown into your GitHub README to link back to this listing. The badge only states the listing — not a security endorsement.

← Back to plugin directory