# Posecode

> Posecode is an open-source text-to-motion language with an Apache-2.0 specification and parser plus an AGPL-3.0 product layer. It is a small, Mermaid-style protocol that LLMs can write to describe single-person movement. A TypeScript parser validates a closed vocabulary and constrains joint targets to configured ranges of motion; a Three.js renderer then animates the result locally in the browser without generative-motion inference.

Key facts:
- A model writes phases and joint angles ("knees: flex 95"), not 3D transforms.
- Authored and IK-generated joint angles are constrained to configured range-of-motion limits; this is not a guarantee that a whole movement is safe or accurate.
- Rendering combines client-side forward kinematics, contact constraints, and inverse kinematics.
- The authoring guide gives an LLM the closed vocabulary, contact rules, and self-checks needed to produce documents that the parser can validate.

## Docs

- [Language specification](https://github.com/posecode-dev/posecode/blob/main/spec/SPEC.md): full grammar, joints, and range-of-motion tables
- [LLM authoring guide](https://github.com/posecode-dev/posecode/blob/main/spec/llm-authoring.md): the system prompt that teaches a model to write .posecode
- [Example movements](https://github.com/posecode-dev/posecode/tree/main/spec/examples): 70+ .posecode documents across fitness, physiotherapy, dance, and more

## Tools

- [Playground](https://www.posecode.org/play): live editor with a 3D viewer, movement library, and shareable links
- [GitHub repository](https://github.com/posecode-dev/posecode): parser, renderer, share codec, embed web component, MCP server, and VS Code extension
- [MCP server](https://github.com/posecode-dev/posecode/tree/main/packages/posecode-mcp): lets any AI agent validate a movement and get a render link

## Optional

- [Landing page](https://www.posecode.org/): project overview and highlights