Currents started with one engineer and one test problem.
In 2019, Andrew was the director of engineering at a startup. His team had trouble scaling their browser tests. He built an open-source tool to fix it and called it sorry-cypress. Soon it helped many companies scale their Cypress tests to millions of test records. In 2021, Andrew turned it into a hosted product: Currents.
The idea was simple. Every test run in CI leaves evidence: errors, steps, traces, screenshots, videos and logs. Most of that evidence disappears when the CI machine shuts down. Currents keeps it. It also joins each result to the history of that test across runs and branches. With that record, you can answer the questions that matter when a test fails:
Cool product, congratulations on the launch🚀
@dunsin Thank you Duns!
The "did this test fail before my change?" bit is the part that stuck with me. That's the first thing I check when a CI run goes red, and it's usually the difference between a real regression and chasing a flaky test. Giving the agent that history instead of just the raw trace feels like the right shape for this.
Congrats on the launch! Gonna try this on our project. We’ve got a ton of unit and e2e tests, and tbh a lot of them are garbage at this point 😅
Any plans to go beyond observability and use the run data to actually help clean up the tests? Like telling us which ones are useless, duplicated, or just flaky noise. That’s the part I’d pay for.
@hungrydog Thanks!
And "a lot of them are garbage at this point" is the most honest description of a test suite I've read today.
Part of this works today, because the plugin gives the agent your run history, not only the latest failure.
Two prompts you can try rightaway:
1. "Use Currents to list the 20 flakiest tests on main in the last 30 days. Read their failed attempts and group them by cause." That's your flaky-noise list, with a reason for each test.
2. "List the slowest tests, and the tests that ran 100+ times in the last 90 days without a single failure." That's your shortlist of tests to cut or merge.
Duplicates aren't in the run data, but the agent has your code too. Ask it to compare the tests on that shortlist and point out the overlap.
One caveat: "never failed" doesn't mean useless. A test that never fails isn't automatically good/bad.
The data gives you candidates, and someone who knows the code makes the call, so ownership is clear.
A proper "clean up my suite" report is exactly the feedback we hoped to get from this launch.
What does "useless" mean for your team: never fails, too slow, or already covered by another test?
I'll take your answer back to the team.
Also, I'd love to onboard you should you require any assistance in using Currents!
The 4 calls and ~1,600 tokens vs 20k for the raw trace is the number that sold me. Most MCP test tools I've seen just dump everything into context. Curious how it handles a monorepo where several Playwright projects report into the same run, does the agent get the failures grouped per project or all mixed together?
@galdayan Grouped, not mixed. Each Playwright project becomes its own group in the Currents run, so several projects that report into one run stay apart.
When the agent reads the run, every failed spec file comes with the project it ran in, and the run details have pass and fail counts per project. Failures with the same error are grouped too, so one broken shared helper doesn't show up as 30 separate problems.
The run summary has no per-project filter, but the agent can open any single spec file or test on its own. And if your packages are separate products, you can give each one its own Currents project, which keeps their test history apart as well.
Seeing whether a test failed before the current change is the detail that stands out. That bit of history can save a detour into fixing code that had nothing to do with the red build.
@oleksii_sekundant Agreed, it's the cheapest check with the biggest payoff. The history answers three different questions:
- Failed on main before your change too: not your PR. Fixing it in your branch is the detour.
- Passed on main, fails only on your branch: most likely your change.
- Flips between pass and fail on main: a flaky test. A real problem, but a separate one from your PR.
But always compare the error, not only the status. The same test failing with a new error on your branch can still be yours.
Today the history is a tool the agent can call, not a fixed step in the skill. If you want it every time, put it in the prompt: "before you change anything, check if this test failed on main in the last 2 weeks."
congrats on the launch sumit. to your question, today its me first, and honestly the agent mostly reads what i copy paste from the log. what i would want is a short list after the run, which failures it fixed alone and which it left for me. i read that before merge, not the whole log. is something like this there?