Programmable AI, from language and model computation to working agents.
The XLang Foundation supports an open-source ecosystem developed by CantorAI and the XLang Foundation. Our current focus brings together three projects: XLang3, Garnet, and xMind.
| Project | Role | Explore |
|---|---|---|
| XLang3 | A native programming language and runtime with Python-compatible syntax, an explicit execution pipeline, and native application integration. | Source |
| Garnet | A programmable, profile-based AI inference runtime and compiler. Model programs stay readable while execution specializes for the model, hardware, precision, and workload. | Source · Website |
| xMind | A native agent platform and coding assistant. CLI, browser, and VS Code clients share a C++ backend for agents, tools, models, sessions, approvals, and execution history. | Source |
Program the application with XLang3. Use Python-compatible source with a native runtime and a pipeline that separates parsing, semantic binding, intermediate representation, and execution. Embed the runtime in native applications and connect scripts with system capabilities.
Program the model with Garnet. Express model computation as backend-neutral xModel programs using XLang3 Tensor Expressions. Garnet captures an inspectable graph, then partitions, lowers, and compiles it for hardware-specialized execution. Profiles describe backend, precision, shapes, and workload constraints. The project includes TensorRT/CUDA, Metal, OpenVINO, native CPU, and custom-kernel work; coverage and validation vary by model and backend.
Build and operate agents with xMind. Start with an agent/tool loop, then add skills, named agents, or graphs as the task requires. xMind embeds XLang3 for compatible Python scripts and pure-Python libraries, while its clients share native session and execution state.
Together, these projects connect application code, model computation, and agent workflows. Each has its own interfaces and validation scope; this ecosystem direction does not imply that every integration or backend is production-ready.
- XLang3 language and runtime
- Garnet inference runtime and compiler · Getting started
- xMind agent platform and coding assistant · Architecture
- XLang — the original language and runtime; XLang3 is developed in its own repository.
- caslang — deterministic agent workflow and tool-calling language.
- Quanta — native C++ vector database work for streaming embeddings and AI applications.
We welcome contributors working on languages, compilers, model execution, native applications, and agent systems. Explore each project's README, architecture, and contribution guidance to find a starting point.
Foundation website · Garnet website · info@xlangfoundation.org