RightNow

Open-source Agent Operating System

Run a 1-billion parameter LLM on a $10 board with 256MB RAM

Autoresearch for GPU kernels. Give it any PyTorch model, go to sleep, wake up to optimized Triton kernels.

An agent harness that compiles a model into one provably-correct, self-retargeting CUDA megakernel and self-tunes it past cuBLAS at batch-1 LLM decode, paper: https://arxiv.org/abs/2606.09682

the agi compiler: records llm agent behavior, proves what repeats, and compiles it into verified, sandboxed wasm binaries that run for microdollars. nothing figured out twice, paper: https://arxiv.org/abs/2607.04542

Pure Triton kernels for Qwen3.5-27B inference on NVIDIA B200

Comprehensive GPU specifications database with 2,824 GPUs across NVIDIA, AMD, and Intel

Claude Code for CUDA. Free AI assistant that actually understands GPU architecture

Minimal TPU implementation with 8x8 systolic array and PyTorch integration

Optimized local serving engine for Kimi-Linear-48B: INT4 quantizer, fused decode kernels for a measured 3.18x, and an OpenAI-compatible server. Ships with k3, a bridge that detects the client per request so Claude Code, Codex, Cline, Aider and opencode all work unchanged.

Dynamic weight generation for recursive transformers via input-conditioned LoRA modulation

Dynamic per-token early exit for LLM inference. Skip layers tokens don't need

Memory-bounded compressed sparse attention via streaming top-k. Triton kernels for the DeepSeek-V4 lightning indexer. 32x regime extension on a single H200 | by RightNow https://www.rightnowai.co/

Open-source transpiler for CUDA Tile (13.1) migration

Single-launch batch-1 decode engine for PrismML Bonsai 27B (ternary and 1-bit) on NVIDIA GPUs. 1.76x the vendor llama.cpp fork on H100, same outputs.

Weights that never freeze. A continual-learning architecture, and the pre-registered experiment that falsified its central claim. https://arxiv.org/abs/2607.20792

GPU CI/CD tool that tests CUDA kernels across multiple GPUs in parallel - Part of RightNow

Open-source web-based GPU performance visualization tool that transforms NVIDIA profiling data into interactive insights for CUDA engineers. Features timeline views, flame graphs, heatmaps, and AI-powered bottleneck detection.

What LLM inference actually costs. 874 verified price rows across 24 providers, 378 GPU rental rates, 395 cited throughput datapoints, and a self-hosting break-even model. Every number carries a source URL, a retrieval date, and passed a mechanical evidence gate. Dated immutable snapshots. CC BY 4.0.

Hierarchical Causal Latent State Machines for Object-Centric World Modeling

Tree execution engine for LLM inference: fork, merge, prune KV cache at token granularity

RightNow Arabic LLM Corpus - One of the largest high-quality Arabic text datasets for LLM training

Agent-native evolutionary optimization. Say the goal in english, evolve code toward a measured target with a population of coding agents.

Faster attention kernels for serving TML's Inkling model on vLLM. 2.7x over the shipping path on H100, and the only implementation that runs on A100.

Official RunInfra SDK (TypeScript + Python) | optimized inference deployments

TypeScript and Python SDKs for the System Zero API

SGLang is a high-performance serving framework for large language models and multimodal models.