Find the error among a thousand lines
95% of a container log is repetitive noise: the same GET /health 200 over and over. The real error —the exception, the stack trace, the dropped connection— is buried somewhere in the middle, and your eyes get tired of scanning.
Radar de logs turns each line into an embedding and computes the file's dominant pattern. The lines that drift from that pattern are almost always the ones that explain the incident. You don't have to tell the model what an error is: it emerges on its own.
Paste your container log and look at the top of the list. That's where the signal is.
Frequently asked questions
Does it detect exceptions without training?+
Yes. It doesn't look for keywords: it weighs what's normal and flags what drifts, whether it's a stack trace, a 500, or a refused connection.
Does it handle multi-line stack traces?+
Yes. Each line is analyzed separately, and stack trace lines usually cluster at the top of the ranking.
What if the error is a single line?+
That's the ideal case: a single line that differs from the dominant pattern scores high on cosine distance.
Related
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.