Changelog

Ask the agent in a Studio chat to run your tests and it does, after asking, on this computer or on the run target you've chosen.

• In the Cloud view too. An Agent session on your project can now run your tests and the scripts it needs, and edit files, asking first. An Ask AI chat (Chat) still only answers. Before, chats in the Cloud view couldn't run anything.
• Claude Code asks instead of refusing. With Claude Code as the engine, a test command shows an approval card, such as "Run "npx playwright test" in your project?", instead of failing with "This command requires approval", from the first message of a new chat. Your approval mode and "don't ask again" choices apply, and a force-push is still refused.
• Runs follow your run target. Each test run the agent starts goes to the Default run target in the composer's session settings or on the Run targets page, the same way a run from Tests does. A target that can't take the run refuses it and says why; the run never moves to this computer instead.
• Claude Code, Codex and Cursor run tests in their own shell, on this computer. When another target is picked, the picker says so.
• Run cards match the run. A card opens for Claude Code's PowerShell runs on Windows and shows only this run's results, never an earlier report. When the engine cut the output short, the card reads "Saw N of M · output cut short" rather than guessing at the rest.

Released as testrelic-cli 1.8.1.

Every Studio chat now runs in Cloud mode unless you turn Local mode on for that chat. Test runs upload, project memory is shared, and a chat in your project still writes and runs your tests under your approval settings.

• Studio opens on the Cloud view. It stays in Code when you launch it on a project with testrelic studio, on the first run while its tools install, or when you've picked Code mode under Settings → Engine & approvals → Default mode for new chats. Moving between views with the header's Code | Cloud switch doesn't change that default.
• Local mode is a Skill you switch on per chat. In a Code chat, open the composer's + menu and pick Skills → Local mode. While it's on, a Local mode chip sits by the input and nothing leaves this computer: test runs aren't uploaded and project memory is read-only. It switches between turns, and the conversation carries on.
• Code chats follow Cloud rules. If your project's TESTRELIC.md describes the project as local-only or upload-free, the agent is told those lines are out of date. Only the Local mode Skill keeps a chat on this computer.
• The terminal CLI is unchanged. Shift+Tab and --mode local still switch modes there. See Modes & engines.

Released as testrelic-cli 1.7.2.

Studio no longer freezes while it reads reports, runs git or starts processes, and it stays quick on projects with thousands of tests and chats with hundreds of turns.

• The window doesn't wait. Disk reads, git, process starts and keychain calls run away from the window, so a click or a keystroke in a terminal or chat never stalls behind them. A big paste arrives whole and in order. When you quit, the window goes at once while Studio finishes closing behind it.
• Start-up doesn't hang on a missing folder. If your last workspace is on an unplugged drive or a network share Studio can't reach, it opens another workspace and says "Couldn't reach your last workspace".
• Tests and Needs rework fill in quickly on large projects, and clicking through the Tests list or holding an arrow key no longer lags.
• Long chats keep up. Answers stream smoothly in chats with hundreds of turns, switching to a long chat shows its newest turns first, and an agent's test run shows its output as it happens.
• Git on busy repos. Source control opens quickly even when Studio's own folders hold thousands of untracked files, and a very large diff opens at its start with Show the full diff for the rest.
• F5 and Ctrl+R no longer reload the installed app, which used to leave every chat's processes running behind it.

Released as testrelic-cli 1.7.2.

Studio now limits what it keeps running and what it keeps on disk. Each limit is on by default and has its own switch.

• Idle chats sleep. A chat you haven't used for 30 minutes stops its assistant, along with the browser tools and memory that come with it, and no more than 5 chats stay awake. Its history stays where it is, and your next message picks the conversation up again. A chat that is working, waiting for your approval or holding a queued message never sleeps. Change the timings, or turn it off, in Settings → Chat → Idle chats.
• Evidence for the last 10 runs. Screenshots, videos and traces are kept for the newest 10 test runs. Older runs keep their reports and results, and a test whose evidence has gone says "Evidence removed: only the last 10 runs keep evidence." Before, the newest 20 runs kept theirs, plus any older run still holding a test's latest result. Switching off Settings → Storage → Keep evidence for the last 10 runs goes back to that rule, but doesn't bring back evidence already removed.
• Cache limits. Files opened from TestRelic Cloud are kept up to 1 GB, and trace frames Studio extracts up to 512 MB, the longest-unopened going first. Each limit has its own switch in Settings → Storage.
• Studio's own folders stay out of your commits. Sync now, Auto-sync and the agent's autopilot leave out Studio's run archive, chat attachments, page screenshots, the SDK's local report, ctrf/ctrf-report.json and the .playwright-mcp snapshots. .testrelic/testrelic-config.json is still committed.

Released as testrelic-cli 1.7.2.

Studio now checks that each project has its TestRelic SDK, and the assistant answers in one language you choose.

• A project without its SDK asks. Opening a project whose framework has a TestRelic SDK that isn't installed shows Install the TestRelic SDK, with the exact command it will run in the project folder: for Playwright, @testrelic/playwright-analytics through the project's own package manager. Install runs it, and if it fails the dialog shows the output with Retry. Not now skips it until the next time you open the project. Frameworks without a TestRelic SDK, such as Cypress, Jest and Vitest, are never asked. Settings → Reporter offers the same Install.
• Replies come in English (US) by default, whatever language your files or results are in. Pick another in Settings → Chat → Reply language; open chats switch from their next turn. A message that asks for a different language still gets one.

Released as testrelic-cli 1.7.2.

• Test runs lists every run the agent starts. A run is filed as soon as it ends. Before, a second run in the same chat turn, or a run that finished after its turn, never appeared. Stopped runs, and commands that aren't test runs, aren't filed.
• Terminals: keys typed quickly reach the shell in order. Hiding the bottom panel keeps every terminal and whatever is running in it, and no shell starts until you first open the panel.
• Live Browser: popups and new tabs identify as Chrome rather than HeadlessChrome from their first request, and a page's native dropdown lists show over the live view.
• Git: the header Git popover closes on Escape, and Source control updates as soon as a sync from the header button finishes.

Released as testrelic-cli 1.7.2.

The default template Ask AI starts each artifact from is now designed for every width, from a share page to the side panel and a phone. Test videos and screenshots embedded in artifacts fit the space they're given.

• A template for every width. In the default template, stat tiles sit four to a row, two in the side panel and one on a phone, and larger cards take the full width. Charts scale with their card, and wide tables scroll inside their own box with the first column kept in view. Slide decks scale rather than reflow.
• Long names wrap. Long test names and URLs wrap in every artifact instead of running past the edge.
• Video fills its frame. An embedded test video takes its height from its width at 16:9, and screenshots at 4:3, instead of sitting in a fixed-height box.
• Screenshots show one at a time, fitted whole, with Previous screenshot and Next screenshot buttons and a counter. Before, the first image was often cropped.
• Embeds keep what the artifact asked for: the right test's recording in a run with several, the moment the video should open at, and any step, log level or search narrowing.
• Cleaner captions. Screenshot captions drop a "Step N:" prefix that never matched the step, and Playwright's failure screenshot reads "At the failure".

See AI Insights & Artifacts.

Settings → Billing has a way to reach the billing team without leaving the page. Any member can use it.

• Billing help in the page header (during a trial, the header shows the trial instead), or Contact billing at the foot of every tab, opens a short form:
  • pick a topic, such as Payment failed, Refund or Tax or billing details;
  • choose the invoice it's about, if it's about one;
  • write your message, and optionally copy someone else.
• Your account details go with it. Sent with your request lists exactly what's included: the workspace, plan, next invoice, billing details, your three most recent invoice numbers, and who sent it from which tab. Your payment method isn't included.
• You get a reference (BH-…) and an email copy. Replies come to your email address. The form says to expect them usually within one business day.

Also in this release, the country and state lists in Billing details open next to their fields again. They had been opening at the top-left corner of the page.