Parallax

Parallax is an open-source, fully decentralized inference engine built by Gradient that lets you pool GPU and CPU resources from multiple machines—regardless of their specs or physical location—to run large language models as if they were on a single supercomputer. Designed for developers, researchers, and AI enthusiasts, Parallax handles model sharding, load balancing, and node discovery automatically, so you can scale inference without buying expensive hardware or relying on centralized cloud providers. Unlike traditional distributed serving frameworks that require homogenous clusters or complex Kubernetes setups, Parallax works on any nodes that can run Python and network with each other. It supports popular models like Qwen, DeepSeek, GLM, MiniMax, Kimi-K2, and OpenAI's gpt-oss, and can leverage both NVIDIA and AMD GPUs via CUDA and OpenClaw integration (as of Feb 2026). The system is fault-tolerant: if a node goes down, inference continues on remaining nodes. Parallax is ideal for privacy-conscious teams that want to run LLMs on their own infrastructure, edge computing scenarios, or hobbyists who want to combine multiple gaming PCs into a powerful inference cluster. It's not a managed service—you deploy and maintain the cluster yourself—but the setup is streamlined with Docker images and an install script. What sets Parallax apart is its focus on decentralization and ease of use: no cloud dependency, no vendor lock-in, and full control over your data. The project is Apache-2.0 licensed, and its community on GitHub has grown rapidly, earning #1 Product of the Day on Product Hunt in October 2025.