The AI SRE that does the work, not just the investigation.

A site reliability engineer keeps software running. They respond when it breaks, fix the cause, and do the upkeep that stops it breaking again. Most AI SRE tools take on the first part: they investigate. bugstack takes on the job. It hears about problems from whatever telemetry you run, fixes them, proves the fix, and looks after your packages, secrets and deploys when nothing is on fire.

What is an AI SRE?

When something breaks, a site reliability engineer gets paged, reads the logs, finds the cause, and fixes it. Between incidents they patch what is vulnerable, update what is stale, and keep deploys healthy. An AI SRE is an AI agent that takes on that work without being asked.

The category is new, and the products in it differ. Most of them, such as Resolve AI, Traversal, Cleric and Datadog’s Bits AI SRE, connect to your logs, metrics and traces, investigate an alert, and give the on-call engineer a likely root cause. Some go on to suggest a fix.

What they promise is a faster investigation. A person still writes the fix, and the upkeep still waits for someone to have time.

The whole job, not one step

Ask what an SRE does in a week, and investigating alerts is a small part of it. bugstack is built for the rest as well.

• It responds to crashes. It reproduces the failure, finds the cause in your code, writes the fix, and proves it.
• It rescues failed deploys. On hosts you have connected. Render is the host supported today.
• It watches for vulnerable packages. Checked every four hours.
• It watches for leaked secrets. Checked every night.
• It keeps packages up to date. Checked every day.
• It leaves a record. Every action is written to a tamper-evident audit log, and every pull request says what happened and why.

bugstack reports what its scans find, and it can open pull requests for those findings. That part is switched on per account. See how it compares with Dependabot and Renovate.

Telemetry agnostic

A fix feature built into a monitoring product works only on that product’s data. bugstack is built to take problems from whatever you already run.

Its own SDK is the quickest way to start. For early access teams we connect Sentry, Datadog or your own source as part of onboarding. You do not have to change monitoring tools, and you do not need one at all.

How bugstack compares

The left column describes the category in general. Individual products differ, so check each vendor’s documentation.

What happens when production breaks

One problem, start to finish:

1. It hears it. A crash arrives from the bugstack SDK, or from the telemetry you already run. Repeated crashes become one problem.
2. It assigns it. A coordinator reads the problem and picks the specialists the job needs.
3. It reproduces it. The tester replays the failure against your code, before any fix exists.
4. It fixes it. The fixer finds the cause in your repository and writes the smallest change that solves it. Anything over 3 files or 30 lines always waits for your review.
5. It proves it. The tester replays the same failure. It has to fail before the fix and pass after. Your own CI runs as well.
6. You decide how it lands. The pull request waits for your review, or merges itself if you have chosen that level.

See how bugstack works in full →

You decide how much it does alone

Every change is a pull request, and your CI runs on it. You pick one of three levels for each repository: "Draft fixes for me", "Open PRs, I merge", or "Fix & merge for me". New accounts start at the level where you do the merging.

Some actions wait for a person at every level: rotating a secret, changing infrastructure, and rolling back a deploy. Undo opens a revert pull request.

Read how bugstack handles your code and data →

What bugstack does not do yet

We would rather tell you now than have you find out later. bugstack does not watch:

• Wrong answers returned as a success. If the code runs and returns the wrong result, bugstack does not see it.
• Anything that fails without throwing an error. A job that quietly stops.
• Slowdowns and performance regressions.
• Errors in your users’ browsers. It works on server-side errors.
• Deploys on hosts that are not connected. Render is the only host today.

bugstack does not replace dashboards, tracing or performance monitoring. Keep those, and let bugstack act on what they report. See how bugstack compares with error monitoring.

How well it works

On 25 September 2026 we ran bugstack against 158 defects planted in real code: our own. Six of seven problem classes reached the right action at least 80% of the time. 45 of 61 crash fixes held, which is 74%. Model spend came to about 25 cents per crash fix.

It is not always right. That is why review is the default. See the numbers →