New features, improvements, and milestones from Quash.
Cancelling one suite step no longer cancels the whole run. Added an offline banner that reliably detects when the Quash backend is unreachable, plus a broad round of performance, bundle-size, and stability fixes.
Re-run reports are now fully AI-authored, with executive summaries and self-healing annotations. Script re-runs can be cancelled end-to-end, and Windows now bundles its own SSH client for cloud tunnels.
Quash MCP lets you run mobile app tests directly from AI agents like Claude Code, Cursor, or Claude Desktop using plain language. It handles test generation, execution, and reports on real devices/emulators (Android + iOS on macOS, Android-only on Windows). One command installs and auto-configures it with your agent.
Read moreTaps in the cloud iOS viewer now route through a native touch path (~1.5 ms vs ~400 ms). Added category and type filters to run prompts, and test-case names now appear as report titles.
Generate, copy, and revoke MCP API tokens, and runs triggered via MCP now appear and live-stream instantly. The app is also ~640 MB smaller after slimming the bundled Python runtime.
Runtime secrets and configuration are now fetched from the backend at launch — no hardcoded keys ship in the app — alongside several startup and security fixes.
Device streams now re-establish correctly when a reloaded window reattaches to a running session, and script-replay timelines rehydrate properly after a page refresh.
Android devices on isolated networks can now be controlled wirelessly through a relay tunnel, and cloud iOS gains H.264 live streaming at device resolution with full gesture support.
Connect to cloud iOS simulators with a live interactive viewer — stream video and send taps, swipes, and system gestures — plus a redesigned checkout with a live price breakdown.
Re-enabled Switch to Wireless for physical Android devices, with clear wired/wireless icons on device cards, and upgraded scrcpy to a universal Mac binary (v4.0).
Added full plan management with a Plans tab, plan-level feature gating, and real-time entitlement updates that apply without a restart, plus an org lock state when a subscription expires.
Each user now manages their own "My Builds" list with favourites and auto-favourite, a per-device build selection step for suites, and App/Build details in reports and PDFs.
Scripted test runs no longer spin up the AI engine — replay is now strictly deterministic — with fixes for heal retrieval, cloud inactivity tracking, and LLM usage billing.
Run a subset of a suite by selecting individual test cases, and override the model, reasoning level, retries, and wait-before per task in the Run Suite modal.
Switched the default model to Claude Sonnet 4.5 and removed the last legacy dependencies from the Mac app — all data now routes through the Java backend — with a 5-minute cloud-device inactivity countdown and stronger connection resilience.
Moved auth, billing, knowledge, integrations, and device management onto a Java backend, and shipped a context-aware script-repair agent plus a real-time event bus for live device and cloud updates.
Web testing gains defined pre/post-condition contracts, saved authentication states, live browser streaming, multi-tab automation, and richer web run reports.
Standalone runs can now pick a test-data row at runtime, new suites can be renamed the moment they're created, and the instruction box auto-resizes.
Redesigned suite run setup as a guided 3-step wizard — device, then test data, then execution mode — with auto-generated trajectory GIFs for script replays.
Re-run any completed task on the same device without calling the LLM. Mahoraga records the execution trajectory and replays it as a self-healing script — adapting to minor UI shifts between builds. Reruns complete in ~1/3 of the original time with zero LLM overhead.
Read moreTest-data slot references now resolve on the server at execution time for both suite and standalone runs, and deferred test-data selection is generally available.
Install a build directly from the Builds tab with a per-device picker, see per-LLM-call cost tracking, and start tasks faster with agent warm-start.
A new engine replays recorded steps with multi-signal element matching and heals automatically when a match fails, alongside a script editor, suite run wizard, and reports analytics dashboard.
Added physical iOS device support and a redesigned test-data selection workflow, plus a full visual rebrand — new icons, font, and amber accent — and org-level feature flags.
Mahoraga comes to the browser. Describe a web flow in plain English and run it the same way you run mobile tests — same reports, same suites, one project for both.
Read moreChrome is now a peer device alongside Android and iOS, driven by a new browser automation agent, with web-aware reports, a redesigned sidebar, and a big startup-performance pass.
Updated the add-funds quick-select amounts to $25 / $50 / $100 and improved auto-update reliability on macOS.
Added browser-agent benchmarks and research artifacts, and re-verified all auto-update artifact hashes against the actual build outputs.
Rewrote the Jira issue picker with infinite scroll, filter pills, and server-side pagination, and corrected LLM cost reporting accuracy.
Mahoraga now runs on iOS simulators. Upload an IPA, write the same plain English instructions, and execute on iOS — same reports, same suites, same workspace.
Read moreForced UTF-8 for all Python subprocess calls to prevent character-encoding errors, and fixed auto-update extraction on macOS and PEM path handling on Windows.
Fixed suite auto-save, enabled run actions for individual test cases, and cleared up prompt-menu issues surfaced right after the 2.0 launch.
A landmark release: a fully redesigned test-suite system, real-time multi-user collaboration, a knowledge service with vector search, native Jira integration, and a visual overhaul across every major view.
Landed private-pool device allocation on the main release branch with automatic iOS keyboard dismissal, plus fixes for stuck stop buttons and private-pool state refresh.
Quash runs natively on Windows 10/11 with full feature parity. Connect devices, run tests locally, and share the same workspace across Mac and Windows.
Read moreThe biggest release yet: full iOS simulator automation goes live, Quash now runs on Windows, organizations get private device pools, and suite reports gain pass/fail charts — plus HubSpot contact sync and long-press gestures.
Introducing Megumi, Quash's dedicated test generation agent inside the new Test Studio. Give it context — a PRD, a Figma file, a GitHub repo, or a plain English prompt — and Megumi generates structured test cases with instructions, expected results, and priority levels. PRD upload — attach a spec and Megumi extracts flows, acceptance criteria, and edge casesFigma integration — Megumi reads component names, states, and annotations from your designsGitHub integration — Megumi generates targeted tests per pull request using real endpoints and field names Works conversationally, remembers the full session, and gets smarter about your app over time through Guidance.
Read moreSimulate 4G, 3G, 2G, or offline conditions on cloud device runs.
Added an iOS tab previewing upcoming simulator support and an attachment dialog with an upload button on the prompt input, alongside cloud-device cleanup and integration fixes.
Connect Slack, Notion, and Jira over OAuth for automated export and test-failure notifications, with end-to-end LLM observability across agent calls and a redesigned cloud-device reservation system.
Try Quash at quashbugs.com/playground — no account, no download. Type a prompt, watch Mahoraga execute on a real app, and see the report.
Read moreA new Devices page in the sidebar brings local and cloud devices into one view, with a 5-minute inactivity countdown that auto-releases idle cloud devices.
A step-by-step wizard for setting up Android emulators, USB devices, and cloud devices, plus a re-run agent that replays recorded actions, validates the final state, and captures screenshots.
The ADB public key for cloud (EC2) devices is now fetched from the cloud and the bundled key refreshed, fixing cloud device connectivity.
Cloud devices reach general availability with a polished queue UI and OS-level notifications, and the macOS app is officially renamed from Mahoraga to Quash, with scrcpy bundled for builds that work out of the box.
Overhauled the video-streaming pipeline for fewer black screens and stalls, added tap/swipe icon overlays to screenshot annotations, and PDF reports now embed the Quash logo and task screenshots.
Rebuilt the suite builder with a full preview panel and per-task stop controls, added server-side PDF report generation, and brought video controls to the trajectory screen.
Verified ADB connections before showing a device as connected, made the inactivity timer more accurate, kept the trial banner always visible, and fixed stuck and zombie device allocations.
Introduced on-demand cloud Android devices — cloud-backed emulators with queue and pool management, live activity tracking, and Slack notifications for device requests.
Deploy Quash's execution layer inside your own AWS, GCP, or Azure environment. Test data and recordings never leave your infrastructure.
Redesigned onboarding, added replay controls in the Device Panel, improved report generation, and added live device-connection monitoring; thinking mode now defaults off.
Quash is now generally available. Free tier at 100 runs/month, Pro with unlimited usage billing, Enterprise with SLAs and dedicated infrastructure.
Read moreNew signups now receive $10 in trial credits (up from $5) and start active immediately — no more waitlist gate — with Slack notifications on successful signup.
AI agents can now trigger test runs, poll results, and retrieve reports via Quash's MCP server. Fully autonomous QA pipelines — no human in the loop.
The first minor release added product analytics, a trial waitlist flow, password reset, trajectory capture on test failures, and per-user referral codes, plus a round of suite and event-system fixes.
Export any execution report as a formatted PDF with optional password protection.
Screenshot understanding is now enabled for every task automatically, and Intel Mac builds correctly import native packages again.
Scoped API keys per integration. Full audit log of every trigger and config change.
New model option in Settings → AI. Stronger reasoning for complex apps with deep navigation and conditional flows.
Added unified search across tasks and suites, raised the default agent step limit to 50 and task timeout to 20 minutes for longer automations, and upgraded the underlying AI models.
Added a full in-app help system and a manual "Check for Updates" button, redesigned the Devices screen, and fixed modal overflow and UI polish across the app.
Downloaded updates now install and apply correctly on next launch, and failed updates no longer retry endlessly and stall the prompt.
Fixed test-suite execution bugs, timeout and conversation-view rendering issues, HTTP status handling, and data-caching inconsistencies.
Corrected ADB binary path resolution that caused device-detection failures on some installs, and fully separated the staging and production update pipelines.
Verify API state during test execution. Define a validation once, reference it with @slug, and Mahoraga fires it inline — catching backend failures the UI won't show.
Read morePer-organization API keys are now encrypted at rest and decrypted only on use, strengthening how your provisioning keys are stored.
Pass/fail trends, flakiness scores, coverage by tag, and slowest tests in one view.
Get notified in Slack when a suite run fails — with a direct link to the report.
Group test cases and run them together — sequential or parallel. Schedule runs and get one unified report.
Mahoraga retries transient failures up to 3 times before marking a test as failed. Flakiness score now tracked per test.
Run up to 20 tests simultaneously. Results aggregate into one report with per-device breakdown.
Write a test in plain English and Mahoraga executes it on a real device. No scripts, no selectors, no framework. Every run produces an AI-written summary, step screenshots, and a session recording.
Read moreTwo new Anthropic models available. Sonnet 4 for speed, Opus 4 for deeper reasoning on complex flows.
Mahoraga is Quash's on-device agent. Auto-installs on Android devices, detects UI elements via accessibility APIs, executes test steps, captures screenshots, and records the full session. Ships with the native macOS app.
Read moreAI-written summary, flagged observations, step-by-step screenshots, and device recording. Shareable via public link.
Shared workspace with apps, builds, credentials, test data, a test cases library, ad-hoc tasks, and role-based access.
Trigger suite runs from GitHub Actions, CircleCI, Vercel, or Jenkins. Pass/fail gating and status badges.
Choose the model powering test generation like GPT-4o, Claude Sonnet, Gemini Pro.
Describe a user flow in plain language and get a structured, editable test case. No setup, no repo, no design files needed.
When UI elements shift between builds, Mahoraga locates the correct element and continues. Healing events logged in the report.
The foundation of Quash. Describe a flow in plain English and Mahoraga handles every interaction autonomously. No selectors, no scripts, no framework. Upload an APK, connect a device, and run.
Read moreReal Android devices in the cloud, available on-demand. Mahoraga pre-installed — connect and start testing in seconds. 60 minutes free.
