Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/nvidia/cuopt/cuopt-installnpx skills add NVIDIA/cuopt --skill cuopt-installgit clone --depth 1 https://github.com/NVIDIA/cuoptWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00040 | $0.01218 |
| Opus 5 | $0.00020 | $0.00609 |
| Sonnet 5 | $0.00008 | $0.00244 |
| Haiku 4.5 | $0.00004 | $0.00122 |
Grade A, and why
cuopt-install scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
curl -s http://localhost:8000/cuopt/health | jq . This is a copy
100% identical to cuopt-install — 0 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
cuOpt Install (user)
Install cuOpt to use it from Python, C, or as a REST server. For building cuOpt from source to contribute or modify it, see cuopt-developer.
System requirements
- GPU: NVIDIA Compute Capability ≥ 7.0 (Volta or newer). Examples: V100, A100, H100, RTX 20xx/30xx/40xx. Not supported: GTX 10xx (Pascal).
- CUDA: 12.x or 13.x. The package CUDA suffix must match the runtime CUDA (e.g.
cuopt-cu12/libcuopt-cu12with CUDA 12). - Driver: NVIDIA driver compatible with the CUDA version.
cuopt-cuXX(Python) depends onlibcuopt-cuXX(C), so installing the Python package also installs the C library and headers. Installinglibcuopt-cuXXon its own does not install the Python API.
Required questions
Ask these if not already clear:
- Interface — Python, C, or REST server? Server can be called from any language via HTTP.
- CUDA version — What is installed? Check with
nvcc --versionornvidia-smi. - Package manager — pip, conda, or Docker preferred?
- Environment — Local machine with GPU, cloud instance, Docker/Kubernetes, or remote/server (no local GPU)?
Python API
Choose one — do not run both. The second install would override the first and can cause CUDA / package mismatch.
pip
- CUDA 13.x:
pip install --extra-index-url=https://pypi.nvidia.com cuopt-cu13 - CUDA 12.x:
pip install --extra-index-url=https://pypi.nvidia.com 'cuopt-cu12==26.2.*'
conda
conda install -c rapidsai -c conda-forge -c nvidia cuopt
Verify
import cuopt
print(cuopt.__version__)
from cuopt import routing
dm = routing.DataModel(n_locations=3, n_fleet=1, n_orders=2)
C API
The C API ships in libcuopt-cuXX, which is also pulled in as a dependency of cuopt-cuXX — so if you already installed the Python package, the C library and headers are already present. Install libcuopt standalone only when you want the C API without Python. Choose one of pip or conda — do not run both.
What ships with it
6 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 133 lines · 40 tokens per session scan A 37940e3e785a
cuopt-install is a skill published in the GitHub repository NVIDIA/cuopt (1,032 stars, last pushed 3d ago), licensed Apache-2.0. It adds 40 tokens to every session and 1,218 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). It is 100% identical to cuopt-install, differing in 0 lines, and is treated as a copy.
Other skills, from other repositories
add-cuda-kernel
Step-by-step tutorial for adding new CUDA kernels to FlashInfer.
debug-cuda-crash
Tutorial for debugging CUDA crashes using API logging.
benchmark-kernel
Guide for benchmarking FlashInfer kernels with CUPTI timing.
cutedsl_megamoe
Skill "cutedsl_megamoe" from flashinfer-ai/flashinfer, covering updating the cutedsl megamoe kernel src, layout, when the kernel team drops a new version of src/ and what not to update here.
perf-torch-cuda-graphs
Apply CUDA Graphs to PyTorch workloads — API selection (torch.compile, PyTorch makegraphedcallables, TE makegraphedcallables, MCore CudaGraphManager, FullCudaGraphWrapper, manual torch.cuda.graph), code compatibility, capture workflows, dynamic pattern handling, and troubleshooting. Triggers: CUDA graph…
perf-optimization-casebook
Casebook of past successful and classic TensorRT-LLM optimizations (runtime/execution and kernel level) recorded as reusable decision precedents. Consult when deciding which optimization to apply for a classified bottleneck or a given config/model/hardware, to find prior art and adapt a proven approach instead of…