LogCrew
Find what broke in a log file. LogCrew groups repeated errors, shows when they started, and can suggest likely root causes with the log lines to back them up.
Features
- Error clusters. The same error with different numbers, IDs, IPs or paths is counted once, with its first and last line. Click a row to jump there.
- Timeline. See when errors began and where the first spike is.
- Stack traces and timestamps. Handles Python, Node and pytest output, ISO and syslog timestamps, and logs with no timestamps at all.
- AI explanation (optional). Up to three likely causes, ranked, each citing line numbers you can click. If the lines it was given don't support a claim, it says "insufficient evidence".
- Reports. Copy the whole thing as Markdown.
Everything except the AI explanation runs locally. No key, no internet.
Usage
Right-click a .log or .txt file and choose LogCrew → Analyze Log File. To analyze part of a file, select the lines and choose LogCrew → Analyze Selection. Both are also in the Command Palette. Results open beside the editor, and line numbers always match the original file.
AI explanation
You use your own OpenAI or Anthropic key.
- Run LogCrew: Set API Key and pick a provider.
- Analyze a log. The AI section shows the lines and tokens it would send, the estimated cost, and how many secrets were redacted. You can expand it to read the exact text.
- Click Analyze with AI. Nothing is sent before that.
A typical run costs a fraction of a cent. The default models are gpt-6-luna (OpenAI) and claude-haiku-5-5 (Anthropic); change them with logcrew.model.
What gets sent. The top error clusters: one representative line each, a few lines of context, and any stack trace. Not the whole file, and no timestamps. Before sending, LogCrew redacts API keys, tokens, JWTs, password= style values, email addresses, and IPv4 addresses (logcrew.redactIPs). This is pattern matching, so skim the preview if your logs hold unusual secrets.
Privacy. Your log stays on your machine. Only the excerpt you approve leaves it, sent with your key straight to the provider. Your key lives in VS Code's secret storage, not in settings or files. There is no backend, account or telemetry.
Cost cap. If the estimate is above logcrew.maxCostPerRunUsd (default $0.10), the button is disabled until you raise the cap or analyze less. Estimates use list prices built into the extension, which can go out of date.
Free and Pro
|
Free |
Pro |
| Clusters, timeline, Markdown copy |
Yes |
Yes |
| AI explanations |
3 |
Unlimited |
| Export report to a file |
No |
Yes |
| Lines per analysis |
First 5,000 |
Up to 50,000 |
You pay your AI provider directly either way. Pro license keys look like LC-XXXX-XXXX-XXXX; enter one with LogCrew: Set License Key. Keys are checked locally.
Settings
| Setting |
Default |
|
logcrew.provider |
openai |
openai or anthropic |
logcrew.model |
empty |
Empty uses the provider's default |
logcrew.maxOutputTokens |
2500 |
Output limit per run, also used in the estimate |
logcrew.maxCostPerRunUsd |
0.10 |
Cost cap per run |
logcrew.redactIPs |
true |
Redact IPv4 addresses before sending |
Limits
- Plain-text logs up to 5 MB / 50,000 lines. Larger files get a message instead of a silent cut. Compressed and binary files aren't supported.
- One analysis per run: no live tailing, no chat.
- LogCrew never edits your log or your code.
- AI answers can be wrong. Check them against the cited lines.
Development
npm install
npm run compile
npm test
npm run package
Press F5 to start an Extension Development Host, then open a file from samples/.
MIT licensed.
