Project Maya

Project Maya is an AI engine runs large mixture-of-experts models on local NVIDIA and AMD GPUs, RAM, and NVMe SSD storage.

Project Maya runs the 321-billion-parameter GLM-5.3-Flash AI model on local hardware. The engine stores active mixture-of-experts on GPUs, secondary experts in system memory, and the remaining experts on an NVMe SSD, moving them dynamically during conversations. It supports OpenAI- and Anthropic-compatible APIs, a browser-based chat interface, image generation, and tool calls. The software includes multiple quantization options such as Maya-S, Maya-S24, Maya-M, and GSQ-RCO 3.5-bit to fit different hardware constraints. It operates entirely on local machines without sending data externally, and provides built-in benchmarking tools.

Project Maya at a glance

Best for: Software engineer
Runs on: Linux, Windows, Web, API, CLI, self-hosted, open source
Pricing: Free
Released: October 2026
Category: AI coding tool
Website: github.com/mw00/project-maya

Key features

Runs 321B mixture-of-experts models locally
Tiered memory management across VRAM, RAM, and SSD
OpenAI- and Anthropic-compatible API server
Browser-based chat interface with image support
Speculative decoding for faster generation speeds
Automated setup and hardware checking script
Built-in performance benchmarking tools

Use it for

Running massive AI models on consumer or workstation GPUs
Serving local AI models via OpenAI- and Anthropic-compatible APIs
Generating text and images privately without external cloud services
Benchmarking local hardware performance for large language models

Alternatives to Project Maya

Strata — Runs large AI models locally on consumer gaming PCs without server reliance.
RunThisLLM — Provides exact hardware and VRAM requirements for over 295 local AI models.
GPUStack — Unifies AI model serving and GPU compute management across heterogeneous hardware.
AirLLM — Runs massive 70B+ parameter models on small 4GB consumer GPUs.
fal — Access 1,000+ generative models and high-performance serverless GPUs.
Lumabri — Runs large AI models across multiple computers by sharing memory and layer workloads.
Imp — Runs local models directly on NVIDIA Blackwell hardware using native 4-bit math.
llama-swap — It runs multiple local AI models with zero external dependencies.
IAI Memory Engine — Provides persistent, local, and encrypted memory for AI agents without APIs.
Fireworks AI — Run optimized open models at lower cost and higher throughput than closed alternatives.
LocalAI — Runs text, voice, and vision workloads on ordinary hardware with an OpenAI-compatible API.
llmrun — Check which open models run on your specific GPU or Mac instantly.

All Project Maya alternatives