Sentry

Promptfree, for everyone, and the only version there is

Copy the prompt for Claude CodeCopy the prompt for CodexCopy the prompt for Cursor

Copy

Do not rebuild this — but note that the open alternative here is unusually good, and self-hosting is a real answer.

Ingest and processing at scale is the product: absorbing a flood of errors during an incident, grouping them, and staying responsive exactly when everything else is on fire. That is what the price buys.

THREE OPTIONS, HONESTLY
- **Keep paying.** Sensible at volume, and the free tier covers a lot of small projects
- **Self-host Sentry.** It is open source, and it is heavy — several services and real memory
- **Self-host GlitchTip**, which speaks the same protocol so every existing SDK points at it unchanged, and is deliberately lighter. That is the pragmatic answer for most people, and it is covered elsewhere in this catalogue

FIRST, LOOK AT THE BILL
- Priced by events. One noisy error in a hot path can consume a month's quota in a day
- Set sampling per project, and rate limits per key. Then fix the noisy error, because it was probably worth fixing anyway
- That audit often removes the reason to move

WHAT MATTERS WHATEVER YOU CHOOSE
- **Grouping is the whole value.** Fingerprint from the exception type plus your own stack frames, ignoring library frames, with the variable parts normalised out of the message. Get it wrong and you have one issue containing everything or ten thousand containing one event each — both unusable
- Source maps uploaded per release from CI and applied at read time, so a map uploaded later still fixes older events. A minified stack trace is no stack trace
- Breadcrumbs — the recent log lines, requests and clicks before the exception — which are usually what actually explains it
- Authorisation headers and cookies redacted, and bodies off unless the project opts in
- Alerts on a new issue, a regression, or a rate threshold, throttled per rule so one bad deploy sends one alert rather than nine hundred
- Retention for events with a sweeper, while issue-level counts survive

AND THE THING TO DO TODAY
Open the issue list and count how many of the top ten you have never acted on. An error tracker nobody reads every morning is a subscription rather than a tool, and the fix is deleting or resolving the noise rather than changing vendor.

THE ONE-LINE VERSION
Audit your event volume first. If you still want out, GlitchTip speaks the same protocol and your SDKs will not notice.