Log Radar
Highlight anomalies in your logs
Paste hundreds of lines from a container. The radar weighs the dominant pattern and pushes whatever breaks the norm to the top. No labels, no supervision, 100% in your browser.
How it works
- 01
Paste your logs
Copy the output of your container (Portainer, Docker, k8s) and paste the text. Nothing is uploaded to any server.
- 02
The model weighs the pattern
Each line is turned into an embedding and a frequency-weighted centroid is computed: the “normal log”.
- 03
The weird stuff rises to the top
Each line is scored by cosine distance to the centroid. The furthest lines are the ones worth looking at.
Use cases
Privacy
Your logs never leave your browser. The model runs locally via WebAssembly; no server, no data transfer, no account.
Frequently asked questions
Are my logs uploaded to a server?+
No. Analysis runs 100% in your browser with transformers.js. The model is downloaded once and all compute is local.
Do I need to label training data?+
No. It's unsupervised: the dominant pattern is computed from frequency and cosine distance alone.
Which model does it use?+
Xenova/all-MiniLM-L6-v2, 22M parameters, quantized to q8. Fast and lightweight for hundreds of short lines.
Does it work with timestamped logs?+
Yes. Even when each line is textually different, the embedding groups the semantically equal ones.
How many lines can I analyze?+
Hundreds without breaking a sweat. The bottleneck isn't the browser — it's pasting more than the eye can review.
Features
- Unsupervised
- No labels, no training. The normal pattern emerges from frequency alone.
- 100% local
- The model runs in your browser. Zero data transfer, zero servers.
- Fast
- 22M quantized parameters. Hundreds of lines in milliseconds after the initial load.
- Free and unlimited
- No accounts, no quotas, no usage limits.
- Stack-agnostic
- Works on any text: logs, build output, CSV, whatever you paste.