eee1723/dsh-houdini
eee1723/dsh-houdini has 123 GitHub stars and 2 forks. GitHot measured -1 stars in 24 hours. Last checked 2026-11-15 15:44 UTC. Last push 2026-11-14.
Measured star growth
- in 24h
- -1
- in 7 days
- Update delayed
- in 15 days
- Update delayed
- in 1 month
- Update delayed
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
About
DeepSeek Harness Desktop 插件,让 AI Agent 用自然语言构建、检查和维护可编辑的 SideFX Houdini 工程,支持程序化建模、HDA、场景诊断与渲染。 | A DeepSeek Harness Desktop plugin that lets AI agents build, inspect, and maintain editable SideFX Houdini projects with natural language, including procedural modeling, HDAs, scene diagnostics, and rendering.
AIREADME.en.md
Originaldsh-houdini
简体中文 | English

Build, inspect, and maintain editable Houdini projects with natural language.
dsh-houdini connects live Houdini scenes to DeepSeek Harness Desktop. Describe your goal in a single conversation, and the Agent can inspect the scene, create and modify nodes, adjust parameters, build tools, and iterate using actual execution results.
The results remain native node networks, control parameters, keyframes, HDAs, and project files. You can review how they were built, edit them manually, or continue refining them through conversation.
Key Capabilities · Quick Start · How It Works · Documentation
Key Capabilities
| Capability | What you can do |
|---|---|
| Scene understanding and diagnostics | Inspect nodes, parameters, connections, and geometry; examine Cook errors and dependencies; use UI screenshots and previews to locate problems. |
| Procedural modeling | Use SOPs, attributes, VEX, and Python to build editable networks, expose control parameters, make local changes, and check geometry. |
| HDAs and artist tools | Package workflows into HDAs; create parameter interfaces, Shelves, Python Panels, Viewer States, and Packages; organize source code, resources, and entry points. |
| Materials, lighting, and rendering | Work with UVs and textures; build MaterialX / Solaris / Karma workflows; configure cameras and lights and generate previews or rendered output; use COP layers and image processing when needed. |
| Parameter controls and animation | Set parameter dependencies and keyframes, organize mechanical hierarchies, rigs, and animation networks, and observe actual output after control values change. |
| Reference-assisted workflows | Work from text, images, reference pictures, and videos; use configured models to generate or edit images, transcribe narration, extract frames, and organize steps backed by source references. |
| Project organization and delivery | Organize node naming, grouping, subnets, outputs, and resource paths; save HIP files, tool source code, and related files for further editing and reuse. |
| Long-running tasks and execution recovery | Observe renders or computations through DSH background jobs, retrieve results for the original request, build and validate in isolated Houdini processes, and inspect execution records and errors. |
These capabilities can be combined: start by inspecting an existing network, turn it into a parameterized asset, and package it as a tool; or start from reference material, build a scene, and prepare a presentation. The Agent chooses the appropriate approach based on your goal, the existing project, and currently available capabilities.
Quick Start
Requirements
| Item | Supported range |
|---|---|
| Operating system | Windows x64 |
| Host | Plugin 0.1.11 requires DSH 0.2.2-rc.1. Users on DSH 0.2.1-rc.5 / rc.3 must explicitly select older plugin version 0.1.6. Confirm the host version before upgrading the plugin. See the compatibility matrix for exact supported combinations and validation boundaries. |
| Houdini | 21.0 / 22.0; see the installation guide for build compatibility information. |
| Models | Configure an available primary chat model in DSH. Image understanding, image generation, and speech features depend on the actual capabilities of the selected models and services. |
Installation
- Install a compatible version of the official DSH Desktop and Houdini, then configure the primary model in DSH's native model settings.
- In Desktop, open Plugins → Add Plugin → Package name or address, enter the exact package specifier or GitHub address below, and click Install. You can also download the
.tgzfrom the release page and enter its local absolute path. - Click Enable Now, or enable
dsh-houdinifrom the plugin list.
[email protected]
https://github.com/eee1723/dsh-houdini/releases/download/v0.1.11/dsh-houdini-0.1.11.tgz
The same complete package is available from npm, npmmirror, and the GitHub release. See the update instructions for package-name and address installations.
Newly installed plugins are disabled by default and provide Houdini mode only after being enabled. The full package includes FFmpeg / ffprobe, and a standard Python interpreter is located automatically from the Houdini installation. Using the plugin requires no separate Node installation, source compilation, or FFmpeg PATH configuration.
See the installation and usage guide for installation, component updates, legacy entry migration, and uninstallation.
Connect to a Scene
| Entry point | How to connect |
|---|---|
| From DSH Desktop | Start or select a conversation in Houdini mode, bind a connected window through the project controls, or choose a Houdini installation to create a scene or open a HIP file. |
| From Houdini | Select DSH-Houdini → Open Workspace to enter the corresponding Houdini conversation in the same Desktop, based on the current project. |
After the plugin is enabled or updated, connection components are installed or updated automatically. New projects reuse the same preparation process, so manual installation is normally unnecessary. If an existing window still has old components loaded, save and reopen that window; if preparation fails, the project menu provides a retry option. Each conversation can explicitly select its own project. Opening another HIP uses a new process by default, preserving existing windows.
Houdini controls appear with the current conversation's Houdini mode, while other modes retain the original DSH interface. For multi-project selection, workspaces, and resuming projects, see the project usage guide.
Houdini mode in Windows Desktop provides an Always on Top button so you can work with the scene and conversation side by side; returning to another mode restores the previous window level. If you need a ChatGPT/Codex subscription, you can install the community plugin [email protected] through the official plugin manager. It is independently authorized and maintained, is not bundled with the Houdini release package, and models are still selected from DSH's native menu.
Usage Examples
Describe your goal, the scope you want to modify, and anything that should be preserved. For example:
Create a parameterized asset
Create a spiral staircase in the current scene, exposing the step count, width, and total height. Keep an editable SOP network and check the relationships between steps after changing parameters.
Inspect and modify an existing network
Find out why my selected node chain has no geometry output. Locate and fix the problem within this chain, preserve the other networks, and explain the changes and validation results.
Build a reusable tool
Package this generator as an HDA, organize its inputs, outputs, and parameter interface, test it in a new instance, and explain the file location and how to load it.
Create materials and a presentation
In this saved project, set up MaterialX materials, Karma lighting, and a camera for the selected asset. Render a preview and check material bindings and rendering errors.
Learn from tutorial material
Analyze the construction method in this video. First organize the key steps, parameters, and source timestamps, then reproduce it in Houdini in stages, marking anything that is unclear in the original video.
Tasks can begin with observation, construction, modification, or troubleshooting. When modifying an existing project, specify the target and scope. For delivery, you can ask the Agent to show key networks, control parameters, actual outputs, and file locations. Evaluate the model's proposals using geometry, visual, and project checks.
How It Works
DSH provides conversation and general Agent capabilities, while the plugin connects the model's operations to real Houdini. The main path for project operations is:
flowchart LR
subgraph desktop["DSH Desktop"]
A["Conversation · Models · Agent tasks"] <--> B["dsh-houdini<br/>Context · Tools · On-demand skills"]
end
subgraph houdini["Houdini"]
C["Bridge<br/>Main-thread execution queue"] <--> D["HOM and operation library<br/>Nodes · Parameters · Geometry · Scene"]
end
B <-->|Local HTTP| C
| Component | Responsibility |
|---|---|
| DSH Desktop | Manages models, accounts, credentials, sessions, workspaces, and general tasks, supporting continuous interaction between the user and Agent. |
| Plugin adapter | Provides Houdini mode, project context, tools, and interface elements; loads domain skills as needed and presents execution results. |
| Houdini Bridge | Receives requests inside Houdini, verifies project identity and operation scope, executes operations through the main-thread queue, and returns actual results. |
Houdini's Python API (HOM) runs only on the Houdini side. Media preparation runs in standard Python worker processes, while isolated builds and validation use one-shot Houdini processes; both are separate from main-thread execution in the current scene.
Execution results, errors, images, and files are returned to the original conversation so you can iterate on actual outcomes. If a request response is lost, the original request can be queried to reduce duplicate operations. DSH tracks long-running jobs, while operations within the same Houdini project remain serialized through the queue.
See the system architecture for detailed responsibilities and source entry points. Tool parameters are documented in the tool guide and operation catalog.
Model and Extension Configuration
The primary chat model, accounts, and credentials are managed in DSH's native settings. Use a model with image input capability for visual understanding; image generation or editing can use a separately configured image service.
Configure speech transcription only when needed. In any mode, go to Settings → Tutorial Videos to select the address source, actual protocol, and exact model, confirm the scheduling settings, and save directly—no need to start a model conversation first. The Agent can also use the same configuration from a conversation. API keys are entered directly into the DSH credential service through a secure input and are not included in chat responses.
Existing DSH providers can be explicitly reused, and standalone speech services can be connected through their actual APIs. Media analysis alone does not require a scene connection. See model and media settings for configuration methods, dependencies, and supported interfaces.
Documentation
| Entry point | Contents |
|---|---|
| Installation and usage | Installation, connection, updates, migration, model configuration, and uninstallation |
| Multiple projects and recovery | Multiple windows, explicit binding, workspaces, and project recovery |
| Tool guide | Tool responsibilities, parameters, and return values |
| Operation catalog | Node, parameter, geometry, asset, observation, and rendering operations |
| Project files and output | HIP files, images, resources, and output directories |
| Execution records and diagnostics | Execution, reference material, and diagnostic information in traces |
| Support reports | Collect, inspect, and export diagnostic material |
| Documentation index | Complete design, interface, and maintenance documentation |
On-demand domain skills
| Domain | Purpose |
|---|---|
| Task decomposition | Organize modules, observations, and integration based on goals and dependencies |
| Isolated execution | Build and validate in an isolated Houdini process |
| SOP | Procedural modeling, curves, and geometry operations |
| External asset integration | Component replacement, materials, instancing, and high/low-poly workflows for generated or imported models; includes offline GLB asset preparation |
| Simulation | Feedback solving, timing, caching, and simulation modifications |
| Surface appearance | Reference images, surface details, UVs, and textures |
| COP | On-demand layer and image processing |
| Tool development | HDAs, scripts, panels, and tool delivery |
| Parameter UI | Control parameters, interface organization, and bindings |
| Network handoff | Node entry points, grouping, notes, and layout |
| Rig and animation | Skeletons, hierarchies, and animation |
| Solaris and Karma | Materials, USD, and rendering |
| Video tutorials | Transcription, visual evidence, and tutorial projects |
| Trace analysis | Inspect requests, tool calls, and failure causes |
| Knowledge maintenance | Maintain domain resources upon explicit request |
Development and Feedback
Build and check from source:
npm install
npm run build
npm run docs:check
npm test
For extension responsibilities, interfaces, and validation methods, see the development guide and architecture code map. Submit issues or suggestions through GitHub Issues. Include your goal, versions, observed behavior, and reproduction steps, and use the support report guide to check what to share.