What is Project Maya?
Project Maya is a local inference engine and server for large mixture-of-experts language models, including GLM-5.3-Flash, DeepSeek-V4.1-Flash-1017 and MiniMax-M3.1-Flash. It keeps frequently used experts in GPU memory, places others in RAM and reads the remaining experts from SSD storage as needed. Its startup process measures available hardware and adjusts resource allocation.
The software includes browser-based chat, a live monitor and a model management interface for downloading, loading and deleting models. OpenAI- and Anthropic-compatible APIs support tool calls and connections to coding agents. Setup downloads the selected model, builds the software and starts a local server. Linux and Windows are listed as supported environments, with Windows and AMD support described as experimental.
Key capabilities
- Tiered experts GPU memory, RAM and NVMe SSD used according to how often experts are needed.
- Self-tuning Measures GPUs, CPU, RAM and SSD at startup.
- Browser interface Chat and a live monitor.
- Model management A Models page that downloads, loads and deletes models.
- Compatible API OpenAI- and Anthropic-compatible, with tool calls for coding agents.
- Concurrency Up to 8 requests answered together.
Where Project Maya fits
- Local serving of large MoE models
- GPU, RAM and SSD expert tiering
- Chat and monitoring in the browser
- Model download and loading
- OpenAI- and Anthropic-compatible endpoints
- Multi-session use
Reviews of Project Maya
Be the first to share your experience with Project Maya.
Write a review
Loading review form…
Alternatives to Project Maya
DwarfStarDwarfStar runs selected DeepSeek, GLM and Qwen models locally, with a conversational CLI, a native coding agent and an HTTP server. It targets specific Metal, CUDA and ROCm hardware configurations.
FreeTokenFreeToken is an Apache-2.0 serving engine for open-weight MoE models on consumer hardware, with a Windows and Linux desktop app, a CLI and compatible APIs.NNeedleNeedle is an on-device model from Cactus Compute for tool calling, schema-constrained extraction and local embeddings on resource-constrained devices.