Ambera

Beside Sentry, or instead of it

A Sentry alternative that comes back after the release.

Sentry tells you the moment something breaks. Forge tells you what the break cost — how many people stopped finishing sign-up — and, a week after you ship the fix, whether the number moved back. If you need the minute, keep Sentry. Forge is the week.

What Sentry is built to answer

Sentry is built for the first minute: an alert the moment an error is new, the stack trace, the release it first appeared in, and a way to mark it resolved so it reopens if it comes back. For a team on call, that is the right tool, and Forge does not try to be it.

What the alert does not tell you is whether the error matters to the product — whether people stopped completing the thing the page is for — or, once the fix is out, whether the week after looks different from the week before.

What Forge reads from the same errors

Your app sends its errors, events and logs to Forge with a small SDK (browser, Node, serverless and Next.js) or plain HTTP. Once a day Forge counts them and states what moved, with the window, the baseline and the rule beside every number.

  • Exact counts. Errors are grouped by a versioned fingerprint and counted from every row, never from a sample.
  • An issue with its rule. An error opens an issue when its events over the last 7 days are at least 50 % above the 7 days before, from at least 20 events. Below that volume Forge says why it is not judging yet.
  • The error beside the journey. When sign-up completion falls on /signup, the issue shows the errors from the same page in the same days, so the two facts are read together.
  • Per thousand sessions. Errors are also stated per 1,000 sessions, so a week with more traffic is not read as a worse one.
  • The release, measured. Your tracker sends the release with every row. Forge compares the days before that version reached production with the days after, under a plan written before it shipped, and says what it saw: improvement, no change or regression.
  • A spec for your agent. Each issue opens with a work spec Forge writes from the counts; on a paid plan the model drafts one too, every number checked against the counts. It goes to Claude Code or Cursor; Forge writes nothing to your repository.
npm install @ambera/forge-tracking

import { createForgeTracker } from '@ambera/forge-tracking'

const forge = createForgeTracker({
  token: 'ftk_…',            // your product's tracking key, safe in page source
  environment: 'production',
  release: COMMIT_SHA,       // how Forge knows which version sent each row
})

forge.track('signup_completed')
forge.captureException(error)

Where Forge stops

A comparison that only lists strengths is an advert. These are the things Sentry does that Forge does not:

  • No pager. Forge reads once a day, over whole days, so it will not wake anyone when production breaks at 3 a.m.
  • No source maps. A stack trace is stored as your app sent it.
  • No performance tracing and no session replay.
  • No suggested fix as a pull request. Forge drafts a spec; merging and deploying stay with your team.

Questions

Can I run Forge beside Sentry?
Yes, and Forge’s own team does: Sentry for the alert in the first minute, Forge for the issue with its evidence and the outcome after the release. The Forge tracker is its own small package and sends to Forge only.
How quickly does Forge open an issue?
Once a day. Forge reads the stored rows every morning (06:00 UTC), over complete days — today, still being written, is never compared — and opens, updates or reopens issues from that read. It is built to say what changed over a week, not to page anyone.
Why would an error that tripled not open an issue?
Because it is under the minimum volume: 3 events against 1 is a tripling and means nothing. Below 20 events in the window Forge withholds the comparison and says so in a sentence, rather than showing a confident number.
What does it cost?
Free for 1 product, no card: the tracking, the issues, the releases and their outcomes, up to 20,000 rows a day. Solo is $29 a month, flat, for 3 products and 100,000 rows a day each. Nothing is metered and there are no credits; the one paid addition to oversight is the model’s draft of a work spec.
What happens to my users’ data?
Person and session ids are hashed together with your product’s id before they are stored, the IP address is never read, and e-mail addresses, IP addresses, query strings and anything shaped like a secret are removed from the texts. Rows are kept 30 days on Free, 90 on Solo, 180 on Team and 365 on Premium. No screen in Forge shows a person, lists the people behind a count, or ranks them.

Send the first row today.

Google, GitHub or an email. 1 product free, no card, no credits.