Warp
Run trillion-parameter models on a laptop by streaming weights from NVMe
Emerged finds new products worth a look, daily — what’s new today
Goethe
Turn a Linux box with NVIDIA GPUs into a private, OpenAI-compatible endpoint
Teams that want models on their own GPUs get one address, scoped keys and crash recovery without stitching SGLang, nginx and key management together themselves.
Goethe installs on a Linux machine with NVIDIA GPUs and serves every loaded model behind one endpoint that speaks the OpenAI and Anthropic APIs, so existing SDKs, LiteLLM and Open WebUI work by changing a base URL. Keys can be scoped, rate-limited and revoked, and you choose whether it listens locally, on a VPN or publicly. It builds on SGLang and adds its own NVFP4 and ternary kernels. Ampere through Blackwell only, recent drivers required. Made by Matteo, Peter and Janos at Mesh Applied.
More like this
Omnara
Durable AI agents you can self-host, on any model and any machine
qwen38-27b-rtx3090
A tuned vLLM stack that serves Qwen3.8-27B fast on one RTX 3090
OpenBot
Self-hosted AI coworkers, each with its own computer and a policy gateway
Phobos
A Rust kernel language that compiles its own GPU code and serves LLMs with it
rGPU
Write PyTorch on your laptop while the CUDA work runs on a remote NVIDIA box