Run every test faster, for less.

Playwright's --shard splits spec files by file order, so every pull request waits on the slowest machine. Currents balances them by duration on the CI machines you already run. Runs finish up to 40% faster, and on spot instances CI cost is up to 70% lower.

Example run

12 spec files · 4 CI machines · chromium

Run timeline balanced by Currents from duration history. All four machines finish between 4 minutes 49 seconds and 4 minutes 52 seconds. Split with --shard, the same run took 7 minutes 41 seconds.

Same spec files, same machines. Only the split changes.

  • Cursor
  • Red Hat
  • Autodesk
  • Snowflake
  • n8n
  • ServiceNow
  • LangChain
  • US EPA
  • Palo Alto Networks
  • Canadian Red Cross

Your CI is only as fast as its slowest shard

A pull request waits for the last machine to finish, and every run pays for the whole suite.

  • Waiting on a pull request

    Days

    How long Planable's pull requests took to merge before Currents. Every push waits for the slowest machine before anyone can trust the result.

    Read the Planable story
  • A CI bill that grows

    Every pull request and every re-run pays for the whole suite again. The more your team ships, the more each month costs.

  • Agent pull requests

    Agents make this worse. Each agent pull request runs the suite again, often several times while the agent iterates, so the cost of a slow, unbalanced run now scales with agent use.

    See how Currents verifies agent work

--shard splits by file order. That leaves machines idle.

Four spec files, two machines. In file order, one machine gets both long files and the run takes 20 minutes. Balanced by duration, the same run takes 13.

--shard · file order

10 + 10 = 20 min · 3 + 2 = 5 min, then idle for 15

Currents Orchestration · by duration

10 + 3 = 13 min · 10 + 2 = 12 min

Two CI machines run four spec files of 10, 10, 3 and 2 minutes. Split with --shard in file order, machine 1 runs spec01 and spec02 for 20 minutes. Machine 2 finishes spec03 and spec04 in 5 minutes and sits idle for 15. Balanced by duration, machine 1 runs spec01 and spec03 in 13 minutes and machine 2 runs spec02 and spec04 in 12. Both splits do the same work on the same two machines. Only the wait changes. The example from the Currents docs. (opens in a new tab)Read the orchestration guide (opens in a new tab)How --shard and workers split a suite on their own: Playwright sharding vs. workers · Playwright CI at scale in GitHub and GitLab

A live queue that balances every run

Currents keeps how long each spec file took in past runs. Each run starts from a queue of those files, longest first, and every machine takes the next one the moment it is free. When one machine runs slow, the others take the files it would have run.

Duration history12 spec files
  • account/profile2m 4s
  • admin/users2m 52s
  • auth/login2m 45s
  • billing/invoices51s
  • cart/checkout47s
  • cart/discounts1m 54s
  • catalog/search2m
  • orders/history52s
  • orders/returns1m 9s
  • settings/team1m 58s
  • shipping/rates54s
  • wishlist/share1m 17s
Live queue4 waiting
  • orders/returns1m 9s
  • shipping/rates54s
  • orders/history52s
  • billing/invoices51s
  • cart/checkout47s
At 2m 52s, machine 1 finishes admin/users.spec.ts and takes orders/returns.spec.ts, the longest file left in the queue. The other three took their next file the moment each came free.
  • On your CI machines

    Your tests run where they run today. Currents hands each machine its next spec file and records the results.

  • Spec files, for Playwright

    Currents balances whole spec files, so one very long file still runs on one machine. Orchestration supports Playwright.

  • In every paid plan

    Orchestration is included in every paid plan. The guide sets it up in your CI config.

    Playwright orchestration docs (opens in a new tab)

Fail fast, re-run only what failed

A run that has already failed shouldn't keep your machines busy. Stop it early, then run again only the tests that broke.

Fail fast at 3 failures, counted on machines 1, 3 and 4. The third lands 2m 52s in, on machine 1. Files already running finish, and 5 never start.
  • Currents counts failed tests across every machine in the run. When they reach the number you set, the run is cancelled on all of them. Spec files already running finish and keep their results.

    Fail fast docs (opens in a new tab)

Rules that keep the pipeline moving

Currents Actions quarantine, skip or tag the tests a rule matches, by file, test title, tag, project, error message, branch and more. Once the Currents fixture is in place, a rule needs no code change and no pull request.

A flaky test stops blocking
Quarantine flaky tests and they still run and report. A quarantined test's failure doesn't fail the run, so the pipeline keeps moving while someone fixes it.
An expiry, if you want one
Give a rule an expiry date, and Currents alerts Slack or email before it lapses.
Only passes count
A quarantined test counts toward usage only when it passes. Skipped tests never count.
See what every rule touched
The Affected Tests tab lists each test a rule quarantined, skipped or tagged.

Ask your agent

Ask

Quarantine the tests that flaked more than 5% this week, until Friday.

Calls
  • currents-get-tests-performance
  • currents-create-action
Answer
2 tests flaked more than 5% this week, at 8% and 6%. Created 2 quarantine rules that expire Fri 20 Nov.

Your agent can create, update, enable, disable and archive rules through MCP when the person who connected it is an Admin or Actions Admin. Every rule sits in the dashboard for the team to review.

A rule steps one known flake aside. Every other failure still has to pass.

Real failures still block the merge

Quarantine takes the noise out of the check and leaves the gate where it was. A real failure still turns the required check red, so green still means safe to merge.

Quarantined · not blocking
The known flake still runs and reports. Its failure no longer fails the check.
Failed · blocking
Any other failure fails Currents Tests, and the merge waits until it passes.

Required in Branch protection · GitHub

GitHub App docs (opens in a new tab)

Run on spot instances without losing tests

Spot instances cost less because the cloud can take them back at short notice. When your CI passes on the cloud's eviction notice, Currents moves that machine's spec files to a healthy one.

The run carries on and every file still reports. It works on orchestrated Playwright runs, on AWS, Azure or Google Cloud.

Read the spot instances guide (opens in a new tab)
Machine 3 is evicted 2m 30s in. Its 2 unfinished spec files move to machines 2 and 4, and all 12 pass in 6m 43s.
  • FundGuard on spot instances, with Currents

    AWS

    CI cost, down 50%

    FundGuard's cost of running Playwright tests on AWS spot instances.

    Azure

    CI cost, down 90%

    FundGuard's cost of running Playwright tests on Azure spot virtual machines.

    Read the FundGuard story

Works with the CI you already run

Each provider has a setup guide in the docs. Orchestration has worked examples for GitHub Actions, GitLab, Jenkins and Nx.

Your tests run on your own machines. Currents decides which machine takes which spec file and records the results.

Five teams published their before and after

Every number comes from a published case study. The label above it says what produced it.

Orchestration and spot instances

  • FundGuard

    Currents Orchestration

    From about 80 to 40min

    FundGuard's end-to-end pipeline, from about 80 to about 40 minutes.

    Spot instances

    CI cost, down 70%

    FundGuard's CI/CD cost, on spot instances with Currents Orchestration.

    Read the FundGuard story
  • Kittl

    Currents Orchestration

    Single-test time, down 50%

    The average execution time of a single test.

    Spot instances

    Compute costs, down 60to80%

    Kittl's compute costs on spot instances.

    Read the Kittl story

Other changes made with Currents

  • Ramp

    Moved to Playwright

    CI fleet, down 62%

    Ramp's CI machine fleet, after it moved to Playwright with Currents.

    Read the Ramp story
  • Sherpany

    Stability work and duration metrics

    From about 18 to 11min

    Sherpany's suite of about 900 tests, after stability work guided by duration metrics.

    Read the Sherpany story
  • Planable

    Ran in parallel · Cypress, 2023

    From 2 hours 45 minutes to 10min

    Planable's end-to-end suite, once it ran in parallel.

    Read the Planable story
“We have a fixed amount of time to make sure a code change does not regress our product. As the cadence of changes grew and the product features expanded this became more of a challenge, especially when taking costs into account.”
Gilad Dayagi / Director of Web Platform Engineering at FundGuard

See what your CI could save

Set four numbers from your own pipeline. The result uses two published figures: runs up to 40% faster with Currents Orchestration, and CI cost up to 70% lower on spot instances.

Less waiting, every monthUp to12,000minEach run up to 8 min sooner.Lower CI cost on spot instances, every monthUp to$840

50 runs × 20 min × 30 days = 30,000 min · 30,000 min × 40% = 12,000 min · 30,000 min × 4 machines × $0.010 × 70% = $840

This is an estimate. The time saved depends on how unbalanced your suite is today. The cost figure assumes your CI machines move to spot instances.

Run it on your own suite (opens in a new tab)

Faster CI without skipping tests

Test selection makes CI faster by running fewer tests. Currents makes the whole suite faster and cheaper to run, so every pull request is checked by all of it.

Skip tests to save time

  • A test selection tool runs the tests it judges affected by a change and skips the rest.
  • A skipped test checks nothing on that pull request. If the selection misses, the break shows up after the merge.
  • That gap matters more when agents write code nobody reads line by line.

Run all of them for less

  • Currents balances spec files across your machines by duration, so runs finish up to 40% faster.
  • Fail fast counts failures across every machine and stops a failing run at the limit you set.
  • Re-run only the failed tests, spread over all your machines.
  • On spot instances, CI cost is up to 70% lower.

How Currents compares with sharding, hosted grids and test selection

Six questions to ask before you change how CI runs. Every answer about another tool is from that vendor's own docs or pricing page, checked in November 2026.

Currents compared with Playwright --shard, Cypress Cloud, HyperExecute and test selection tools, on six criteria (scrolls sideways on small screens)
CriterionCurrentsPlaywright --shardBuilt into PlaywrightCypress CloudHyperExecuteTestMu AI, formerly LambdaTestTest selection toolsDatadog Test Impact Analysis, CloudBees Smart Tests
Runs on your own CI machinesYesYesYesOn its own cloud VMs, or a private cloudYes
Balances machines by duration historySpec files, from a live queueSplits by file or test countSpec files, by past durationsSplits using past runs and statusesDatadog: selects tests, no splittingCloudBees: bins by past duration
Spot instance supportMoves spec files off a machine the cloud evictsNot documentedNot documentedNot documentedNot documented
Quarantine or skip without a code changeQuarantine, skip or tag by ruleSkips are set in test code or configNo built-in quarantineTest Muting skips a test, by hand or after repeated failuresDatadog: quarantine or disableCloudBees: quarantine
FrameworksPlaywrightCypress up to 12.17Others' results as JUnit XMLPlaywrightCypressSelenium, Playwright, Cypress, Appium and moreMany, per language: Jest, Vitest, Playwright and more
Pricing modelPer test resultOrchestration in every paid planFree and open sourcePer test resultPer parallel testDatadog: per active committerCloudBees: custom quote
Sources for the other products' cells

Test it against the CI you run now

Run your suite through Currents on the same machines, then compare the run time with your old runs. The numbers come from your own pipeline.

Start a 14-day free trial (opens in a new tab)

Three steps to a faster pipeline

Minutes from signup to your first recorded run.

  1. Record every run

    Start a trial, then add the reporter and your project's keys to CI.

    Start a 14-day free trialRead the setup guide
    Prompt for your agent
    Set up this repository to report test results to Currents. Read https://docs.currents.dev/getting-started/setup-with-an-agent.md and follow it.
  2. Let Currents split the run

    Swap --shard for one command. Currents balances spec files by duration on your own CI machines. Included in every paid plan.

    Read the orchestration guide
    CI test step
    # Before
    npx playwright test --shard=${{ matrix.shard }}/${{ strategy.job-total }}
    
    # After
    npx pwc-p run
  3. Gate the merge

    Install the GitHub App and make the Currents check required. GitLab and Bitbucket too.

    GitHub App docs
    Pull request

Questions teams ask

Stop waiting on your slowest shard