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:
Hi Product Hunt!
I'm Sumit from Currents.
Currents records every Playwright test run in your CI: the errors, the steps, the traces, screenshots and videos, and the history of each test.
Until now, people read that record in our dashboard.
Now Claude and Codex can read it too.
Why we built this
I watched the same loop happen again and again this year. An agent writes the code and runs the tests on its machine. It says "done". Then CI goes red on the PR, and the agent doesn't know, because it can't see CI. So a person opens the dashboard, finds the failed test and pastes the error back into the chat. The agent guesses a fix, and the loop starts again.
Writing code got fast. Proving that the code works did not. The person who copies errors from CI into the chat became the slowest step.
But the odd part was: Currents already had the answers. Did this test fail before this change? Does it fail only in CI? Did other tests fail with the same error? It was all in the record. But the record lived in a dashboard, and agents don't open dashboards.
So we spent the last few months on the agent side of Currents. We shipped a hosted MCP server with OAuth sign-in. We shipped a Trace API, so an agent can read a Playwright trace one piece at a time. In our launch example, that took 4 calls and about 1,600 tokens. Every file in that trace comes to about 20,000.
The last gap was setup. An MCP server helps nobody if people can't find it, or if they first have to copy a server URL and an API key into a config file. So we listed Currents in the Claude directory and in the ChatGPT/Codex plugin directory. You find it, click connect and sign in.
What you can ask
"What broke in the last failed run on main? Open a Jira issue for it."
"Which tests were flakiest on main this week, and why?"
"Is this pull request safe to merge? Compare its runs with main."
The plugin also brings three skills:
fix-failing-tests finds the failed run, reads each failure with its steps and trace, fixes the cause, runs the test again and asks before it commits.
collect-evidence finds the run before your change and the run after it, and writes up both with links.
browser-evidence (experimental) records the browser before and after a change and shares the link.
You act as yourself, so your Currents role decides what the agent can do.
Who it's for
Teams that run Playwright in CI and work with Claude, Claude Code, ChatGPT or Codex. The Claude listing works on every Claude plan, including Free. Currents has a 14-day free trial, and you don't need a credit card.
What I'd love from you
Try it, and tell me where it broke or where it confused you. I also have a question for you: when an agent's PR fails in CI today, who reads the failure first, you or the agent?
I'll answer every comment today, and so will the team.
Claude: https://claude.ai/directory/currents
ChatGPT and Codex: Open Plugins and search for "Currents", click connect and just sign in!
Docs: https://docs.currents.dev/ai/connect
Also, if you are participating in the puzzle, just drop your answers down in the comments.
See you in the dashboard, or your nearest AI agent!
what does it do when the test itself is out of date?
@leoheo Good question! It updates the test. The rule the agent follows is "change the code or the test, whichever is wrong", and keep the change to what the failure needs.
To tell which one is wrong, it reads what Currents recorded at the failure: the error, the steps before it, and the page as it was at that moment. Then it compares that with your code changes.
Example: a PR renames "Sign in" to "Log in". The test fails because it can't find "Sign in". The page snapshot shows a "Log in" button and the diff shows the rename, so the agent fixes the test, not the app.
When it's less clear, like the app now shows a different total, the report says which side it changed and why. It also doesn't commit anything unless you ask, so you make the final call.
when the agent reads the failure does it ever just retry until green instead of fixing anything
@igorblink Good question, and yes, agents do this. When all an agent has is "CI is red", a rerun or a bigger timeout is the cheapest guess that turns it green.
Two things push against that with Currents:
1. The fix-failing-tests skill the agent reads first has an explicit rule: fix the cause, not the symptom. No adding retries, raising timeouts, or skipping or quarantining a test unless you ask for it. If the cause is outside the code (the environment, a third-party service, test data), it should say so instead of editing the test.
2. It doesn't rerun to find out what failed. It reads the run Currents already recorded: the error, the steps before it, the trace and screenshots. For a test that passed on retry, it reads the failed attempt too, and has to say whether that flake is worth fixing now and why.
I've had many failed CI runs on main hold up deploys this month, and honestly a lot of the time the Agent reads the failed workflow before I do! Question: a job that times out with every test passing... does that show up as a failure? Congrats on the launch
Cool product, congratulations on the launch🚀
@dunsin Thank you Duns!
Congrats on the launch! Looks like a promising product! 🎉👍
@madzadev Thank you so much!
I'd love for you to try it out!
Congrats on the launch @sumitsaurabh927 and the @Currents team! 🎉
@camelcaseguy Thanks a ton! 🚀