# RightNow-AI/autokernel

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

Repository: https://github.com/RightNow-AI/autokernel
Canonical: https://ross.abutalabs.com/products/autokernel
Homepage: https://www.rightnowai.co/forge
Language: Python
License: MIT
License Family: permissive
Topics: autoresearch, cuda, gpu, kernel-optimization, pytorch, triton
Last push: 2026-03-19T03:13:43+00:00

## Health v2 (maintenance only)
Score: 48/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 73, release rhythm 35, longevity 12
- inputs: {"age_days": 176, "days_push": 167, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1535, forks 161 (observed 2026-08-28T04:04:59.759073+00:00)

## What it is
AutoKernel is an open-source autoresearch pipeline that takes any PyTorch model, profiles it to find GPU kernel bottlenecks, extracts them as Triton or CUDA C++ kernels, and lets an AI coding agent autonomously optimize them via an edit-benchmark-keep/revert loop. It ships with profiling, extraction, benchmarking, and verification scripts plus a comprehensive program.md that guides agents through hours of unattended optimization.

## Use cases
- speed up my pytorch model with custom triton kernels
- automatically optimize gpu kernels overnight
- find which kernels are bottlenecks in my llama model
- generate cuda kernels faster than torch.compile
- benchmark and verify kernel correctness automatically
- reduce gpu inference costs for my model
- extract bottleneck ops from a pytorch model into standalone kernels

## When to choose
- you have an NVIDIA GPU (H100/A100/RTX 4090) and a PyTorch model that is too slow
- you want autonomous, unattended kernel optimization driven by a coding agent like Claude or Codex
- you want verified correctness checks and roofline analysis alongside speedups
- you prefer an open-source MIT-licensed pipeline you can inspect and customize

## When to avoid
- you have no NVIDIA GPU or use AMD/Apple silicon
- you need a fully managed, enterprise-supported drop-in kernel replacement service (the vendor's Forge product targets that)
- your model is tiny or already well-optimized so kernel-level gains are negligible
- you cannot run an external LLM coding agent in your environment

## Facets
- artifact type: cli-tool
- maturity: active
- function: machine-learning, benchmarking, gpu-computing, llm-inference, agent-framework, developer-tools
- domain: machine-learning, gpu-computing, performance, developer-tools, deep-learning
- platform: python, cli
- tags: triton, cuda-kernels, kernel-optimization, autonomous-agents, pytorch, autoresearch, profiling, amdahls-law, linux, gpu

## Member repositories
- RightNow-AI/autokernel (main) score 48

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:59.759073+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:31:07.761638+00:00, confidence not recorded.
  - readme: https://github.com/RightNow-AI/autokernel (fetched 2026-08-28T04:04:59.759073+00:00, sha dc3df724427e)
  - homepage: https://www.rightnowai.co/forge (fetched 2026-08-29T11:33:07.867893+00:00, sha 623f618c156c)
  - site_page: https://www.rightnowai.co/docs (fetched 2026-08-29T11:33:07.879376+00:00, sha 66a23f9f0969)
  - site_page: https://www.rightnowai.co/changelog (fetched 2026-08-29T11:33:07.877234+00:00, sha 1532de1807ce)
  - site_page: https://www.rightnowai.co/pricing (fetched 2026-08-29T11:33:07.881233+00:00, sha 93aa309811e4)
- Data as of 2026-08-30T08:39:29.467469+00:00.