MLC AI
MLC is an open community of machine-learning compilation projects working with the broader ML-systems ecosystem to make model deployment accessible across cloud and edge. Its stated goal is to democratize efficient deployment with open-source tools, frameworks, and best practices across diverse hardware.
At a Glance
- ML researchers and compiler developers
- Application developers building local, browser, mobile, and edge AI
- Cloud and infrastructure teams serving open-source LLMs
- Universities, research labs, and open-source ML communities
- +1 more
AI Tools by MLC AI
(3)
MLC LLM
Open Source LLM Compiler Engine
XGrammar
LLM Structured Generation Library
WebLLM
In-Browser LLM Inference Engine
Discussions
No discussions yet
Be the first to start a discussion about MLC AI
Latest News
KCoral: Lightweight Benchmark Server for Agentic GPU Programming
TIRx Harness: An Open Compiler Harness for Agentic GPU Programming
PithTrain: A Compact, Agent-Native MoE Training System
XGrammar-2: Fast and Customizable Structured Generation for Tool Calling and Agents
Products & Services
Machine-learning compiler and high-performance universal LLM deployment engine. It compiles models for CPUs, GPUs, and mobile devices and exposes a unified engine through REST, Python, JavaScript, iOS, and Android APIs.
In-browser LLM inference engine using WebGPU, with zero server dependency, client-side privacy, offline capability, OpenAI-compatible APIs, streaming, JSON mode/function calling, and custom MLC-format model support.
Open-source constrained-decoding library for JSON, SQL, context-free grammars, and other structured outputs, with C++, Python, and TypeScript interfaces.
Structured-generation system for tool calling and agent applications, adding structural tags, cross-grammar caching, repetition-state compression, and serving/speculative-decoding support.
Market Position
MLC differentiates itself as an open, compiler-first deployment stack that targets many hardware and client platforms from one codebase, including browser and mobile environments. Its closest alternatives or adjacent competitors include llama.cpp, vLLM, NVIDIA TensorRT-LLM, ONNX Runtime, PyTorch/TorchInductor, and browser-focused runtimes such as Transformers.js; MLC emphasizes compilation portability, MLCEngine API unification, WebGPU client-side inference, and structured-generation integration.
Leadership
Founders
Tianqi Chen
Associate Professor in machine learning and computer science at Carnegie Mellon University and Distinguished Engineer at NVIDIA; formerly Chief Technologist at OctoAI. His official profile identifies MLC-LLM, Apache TVM, and XGBoost among the open-source systems he started.
Executive Team
Tianqi Chen
MLC-LLM creator; CMU Associate Professor and NVIDIA Distinguished Engineer
Machine-learning systems researcher; formerly Chief Technologist at OctoAI; creator/starter of MLC-LLM, Apache TVM, and XGBoost.
Board of Directors
Founding Story
MLC grew from an open-source ML-compilation effort aimed at bridging high-level ML models and diverse hardware backends. The initial MLC-LLM repository commit was made on April 28, 2023, and the project vision is to let developers compile, optimize, and deploy models natively on many platforms rather than depend on a single cloud or GPU stack.
Business Model
Revenue Model
MLC presents itself as an open-source community and publishes Apache-licensed software, documentation, and model-deployment tooling. The reviewed official materials describe community contributions and supporting organizations rather than a paid subscription, hosted API, or commercial license model.
Target Markets
- ML researchers and compiler developers
- Application developers building local, browser, mobile, and edge AI
- Cloud and infrastructure teams serving open-source LLMs
- Universities, research labs, and open-source ML communities
- Agent and tool-calling application developers
- Deploying open LLMs on cloud servers and heterogeneous GPUs
- Private, offline, or low-cost AI assistants running in browsers and on consumer devices
- Production LLM serving, multi-GPU inference, and prefill/decode orchestration
- Reliable JSON, SQL, schema-constrained, and tool-calling generation
- Research and development of optimized ML kernels, compilers, and agentic GPU programs
- vLLM
- SGLang
- NVIDIA TensorRT-LLM
- Modular MAX