1Install
One command into ~/.local/bin; no root.
$ curl -fsSL https://autooptm.com/install.sh | sh
Free speedups · 50 credits at sign-up→
Make your code
Make your code
run faster
Run autooptm in your project: the agent measures, changes and verifies on your own machine, and hands you a patch.
$ curl -fsSL https://autooptm.com/install.sh | sh $ cd your-project && autooptm
- 50 credits at sign-up
- No machine charge on yours
- Cloud estimate free
- First cloud run free
- No speedup, deposit back
Linux / macOS · How it works → · Gift rules →
No suitable GPU, or rather leave it to us? Paste a repo or upload a zip; the estimate is free:
before
937.8ms
→
after
249.2ms
same command · same inputs · same output
Measured on public repositories · end to end · output verified
23.41×csvkit
14.65×networkx
13.15×SwinIR
9.16×xarray
8.56×CLIP
7.59×CycleGAN / pix2pix
7.48×litgpt
5.82×open_clip · CLAP
5.24×statsforecast
5.22×DINOv2
4.90×BasicSR
4.58×coqui TTS
3.80×gated_attention
3.75×sapiens
3.69×Wan 2.1 VAE
3.44×vision · ResNet-50
3.42×gaussian-splatting
3.41×mmaction2 · TSN
3.37×sapiens2
3.35×Real-ESRGAN
3.31×OpenPCDet
3.29×PaddleOCR
3.29×CUT3R
3.03×sentence-transformers
2.68×moviepy
2.65×pytorch-image-models
2.62×yolov5 · detect
2.61×shallow-vs-deep-alignment
2.53×DeepSpeed · inference
2.48×nerfstudio
2.26×transformers · GPT-2
2.25×DiT
2.23×YOLOv12
2.13×Wan 2.2 VAE
1.98×LightningDiT
1.90×CoTracker
1.86×PySceneDetect
1.84×GroundingDINO
1.82×SAM 2
1.77×stable-diffusion-webui
1.65×REPA-E
1.64×HunyuanVideo VAE
1.63×RIFE
1.63×SAM 3
1.58×vggt
1.56×WhisperLive
1.56×fast3r
1.49×bd3lms
1.47×hifi-gan
1.41×nanoGPT
1.41×MiniMax-H3 VAE
1.39×GFPGAN
1.35×VibeVoice
1.28×TabICL
1.24×yolov5 · train
1.22×mmdetection · Faster R-CNN
1.22×nwm
1.19×whisperX
1.18×darts
23.41×csvkit
14.65×networkx
13.15×SwinIR
9.16×xarray
8.56×CLIP
7.59×CycleGAN / pix2pix
7.48×litgpt
5.82×open_clip · CLAP
5.24×statsforecast
5.22×DINOv2
4.90×BasicSR
4.58×coqui TTS
3.80×gated_attention
3.75×sapiens
3.69×Wan 2.1 VAE
3.44×vision · ResNet-50
3.42×gaussian-splatting
3.41×mmaction2 · TSN
3.37×sapiens2
3.35×Real-ESRGAN
3.31×OpenPCDet
3.29×PaddleOCR
3.29×CUT3R
3.03×sentence-transformers
2.68×moviepy
2.65×pytorch-image-models
2.62×yolov5 · detect
2.61×shallow-vs-deep-alignment
2.53×DeepSpeed · inference
2.48×nerfstudio
2.26×transformers · GPT-2
2.25×DiT
2.23×YOLOv12
2.13×Wan 2.2 VAE
1.98×LightningDiT
1.90×CoTracker
1.86×PySceneDetect
1.84×GroundingDINO
1.82×SAM 2
1.77×stable-diffusion-webui
1.65×REPA-E
1.64×HunyuanVideo VAE
1.63×RIFE
1.63×SAM 3
1.58×vggt
1.56×WhisperLive
1.56×fast3r
1.49×bd3lms
1.47×hifi-gan
1.41×nanoGPT
1.41×MiniMax-H3 VAE
1.39×GFPGAN
1.35×VibeVoice
1.28×TabICL
1.24×yolov5 · train
1.22×mmdetection · Faster R-CNN
1.22×nwm
1.19×whisperX
1.18×darts
Same command, timed end to end, output checked. Results vary with the GPU and the inputs.
01 Getting started
Three steps to a free speedup
2Open it in your project
In the environment your program runs in (conda / venv). The first time, you approve this machine in the browser.
$ cd your-project && autooptm3Type the command, press enter
The line you normally run it with. The agent optimises, and you end with a patch.
> python train.py --epochs 1 $ git apply autooptm-*.patch
Or not on your machine at all: upload the code, or autooptm run --cloud ., and it runs on our GPUs; the estimate is free. Full docs →
A patch
$ git apply autooptm.patch $ python your/entry.py …
A standard diff with no dependency on our service. Your code runs exactly as it did before.
A number
3.75×
Your command and your inputs, sampled repeatedly before and after, with the median reported.
03 Why us
| AutoOptm | ClaudeOpenAICodexQwenDeepSeekPrompting an LLM yourself | Hiring an engineering team | |
|---|---|---|---|
| Speedup | 2.48×median · 50+ projects measured | ~1.4×typical | Variesdepends on who and how long |
| Cost | $3.5a third of runs cost this or less · median $4.60 · free if no speedup | $9+coding plan + rented GPU · ~$19 on API pricing | $4k–8k~$100/hour × 1–2 weeks |
| Time to result | 1–3 h2.3 h median · unattended | 1–2 hyour own time, prompting and watching | 1–2 wkscheduling, meetings, hand-off |
| Your time | Submit and walk away | All of it | Meetings, Q&A, review |
| Output changes | Never; out-of-tolerance changes revert | Unchecked | Up to the engineer |
| Proof | Frozen harness, median of repeated runs | Build it yourself | Self-reported |
| No speedup | Not charged | Tokens still billed | Hours still billed |
04 Integrations
Your own machine
recommended
$ curl -fsSL https://autooptm.com/install.sh | sh $ cd your-project && autooptm
Your code runs only on your machine; the agent works remotely. No machine charge; tokens at 10% of list, billed as used.
Our GPUs
$ curl -fsSL https://autooptm.com/install.sh | sh $ autooptm run --cloud . --entrypoint infer.py $ autooptm unlock <jobId> $ git apply autooptm.patch
A free estimate first, then it runs to the end and prints the speedup and price. You review the file list before anything uploads.
run --cloudestimatedecidestatusunlockcancelbalancedatasetskeysloginlogoutinstall-skill
Agent skill
$ curl -fsSL https://autooptm.com/install.sh | sh $ autooptm install-skill
Then just tell Claude Code: "optimise this repo with AutoOptm".
05 Who this is for
ML engineer at a tech company
½the compute, same quality
Halve the step time; run twice the experiments on the same budget.
Grad student
2×experiments per week
Twice as fast on one GPU is a second GPU for free.
Indie developer
2×the QPS, same GPU
The same 4090 serves twice the requests at half the unit cost.
Competitive ML practitioner
+3more models per night
They train three models a night; you train six.
06 Code security
Your code is deleted when the run ends
Runs only in a disposable sandbox
One isolated sandbox per run, destroyed after; our dispatcher never runs your code.
No repository access
No tokens, no GitHub App. Paste a public URL or upload a .zip.
Results are yours alone
Reports and patches need sign-in to download; your code and data never train models.
The full terms are in the privacy policy.
07 Pricing
Metered on actual usage
Quoted first, charged after
0Free estimate
2 creditsContinue: deposit
quote − 2Unlock the patch
No speedup, or the failure is ours: the deposit comes back. Refund details →
New accounts get 50 credits: ready to spend on your own machine in the TUI, and on our GPUs too while you subscribe. Invite a friend: when they make their first purchase you get 10 credits, and once they have paid $28 you each get 20. Gift and referral rules →
Trial pack
$3
10 credits · $0.30 / credit
- The best rate we offer, once per account
- ≈ 2–3 full runs on your own repository (a run takes about 2–5 credits)
- Credits valid 24 months
- Private code supported (zip upload)
Usage pack
$5 +
$0.77 / credit · +30% in beta · any amount
- Credits valid for 24 months
- Private repos, results kept private
- Buy more as needed, no commitment
- A run takes about 2–5 credits ≈ $1.5–4, depending on code size, complexity, what the run finds and the speedup
- Card or WeChat Pay, any amount; WeChat Pay requires a QR scan on desktop
$0.50–$5000. Card or WeChat Pay; WeChat Pay requires scanning a QR code on desktop.
Pro +50% credits · beta
$28 / mo
42 credits · $0.67 / credit
- 50% bonus credits per dollar during the beta
- ≈ 10–15 full runs per month
- Private repos · 5 concurrent
Max +60% credits · beta
$99 / mo
158.4 credits · $0.63 / credit
- ≈ 40–60 full runs per month
- Private repos · priority queue · 5 concurrent
Max+ +70% · beta
$199 / mo
338.3 credits · $0.59 / credit
- ≈ 85–130 full runs per month
- Top queue priority · 5 concurrent
- For heavy individual use: large repos, many parallel runs
Beta bonus · top-ups +30% · Pro +50% · Max +60% · Max+ +70%
Subscription credits reset each period · packs last 24 months
Packs: WeChat Pay (desktop QR) or card · Subscriptions: card only · WeChat users can buy the 30-day pack at the same price and bonus
See how much faster your code can run, free
Install it and run autooptm in your project; the 50 credits you get at sign-up are ready to spend. Rather not run it here? Hand the code to our GPUs; the estimate is free.