GitHub - tonis2/rigsmith: Rigging and animating studio

Rig and animate 3D characters. Open a .glb, place a skeleton, skin it, pose it and key it, or generate motion from a text prompt. Everything is saved back into the same file. Rigsmith runs on macOS, Windows and Linux through Vulkan.

demo.mp4

Download

Get the latest build from Releases:

Platform	File	How to run it
Linux	`Rigsmith-<version>-x86_64.AppImage`	Make it executable (`chmod +x`), then double-click it.
Windows	`Rigsmith-<version>-windows-x64.zip`	Portable: unzip anywhere and run `rigsmith.exe`.
macOS 26+ (Apple Silicon)	`Rigsmith-<version>-macos-arm64.dmg`	Drag Rigsmith to Applications. The app isn't notarized, so the first time, right-click it and choose Open.

The first time it starts, Rigsmith copies its built-in animations to Documents/Rigsmith/animations. They show up in the asset browser, and that is where Open starts.

On Windows, if rigsmith.exe won't start, install the Microsoft Visual C++ Redistributable.

Features

Rigging: place joints by hand, or fit a biped, quadruped or bird body plan to the mesh.
Joint limits: keep joints inside natural ranges while you pose.
Skin weights: simple automatic weights, which you can view and paint.
Animation player: a timeline with looping, playback speed and additive layers.
Asset browser: your folders of .glb clips and body plans, searchable, with a built-in library to start from.
Poses and keyframes: key poses, auto-key, saved poses, and tweening between them.
Text to motion with Kimodo: see below.
Works with AI agents: a built-in MCP server lets an agent such as Claude rig, pose and animate alongside you.

Text to motion (Kimodo)

Rigsmith runs NVIDIA's Kimodo motion model on your GPU through Vulkan, so text-to-motion works on macOS, Windows and Linux. There's no Python or CUDA to install, only the model files.

Download the weights from Hugging Face:
- Text encoder: LocalAI-io/Llama-3-Kimodo-GGML, the Llama-3-Kimodo-*.gguf file and its tokenizer.gguf.
- Motion model: LocalAI-io/Kimodo-SOMA-RP-v1.1-GGML. NVIDIA's original folder, nvidia/Kimodo-SOMA-RP-v1.1 with its model.safetensors, also works.
Put them in any folder, for example ~/models/kimodo/.
In Rigsmith, open File › Settings › AI. Choose the text model .gguf, then the motion model (its .gguf, or the NVIDIA folder). Only set the tokenizer if it isn't next to the text model.
Open the Generate tab, describe the motion, and generate.

The weights have their own licenses: the NVIDIA Open Model License and the Llama 3 license.

Using Rigsmith with AI agents (MCP)

Rigsmith serves its tools to MCP clients while it's open, on http://127.0.0.1:8808/mcp by default (only your own machine can reach it). To connect an agent:

Open File › Settings. The MCP section shows the server's address.
Click Copy client config and paste it into your agent's MCP settings.

You can change the port there, and the change applies straight away. Setting it to 0 turns the server off.

Building from source

You need c3c 0.8.3, Git LFS for the sample models, and a Vulkan driver. Windows and Linux use the system's driver; on macOS, setup.sh fetches one.

git clone --recursive https://github.com/tonis2/rigsmith.git
cd rigsmith
git lfs pull
libs/three/setup.sh          # submodules, plus the macOS Vulkan driver
c3c build rigsmith
./build/rigsmith assets/animations/humanoid.glb

Tests: c3c test --test-noleak (fast); c3c test (also checks for memory leaks).
Packages: after a build, packaging/package-{macos,linux,windows}.sh produce the release files. Pushing a v* tag builds them all on GitHub Actions.

License

See LICENSE.