"Show HN: Deltix – AI Driven Testing" is the post a developer going by oneounceman put up on Hacker News on August 15, 2026, describing a tool where an AI agent runs a plain-English task against your iOS app in a simulator and reports back whether a real user could finish it. By the next morning it had picked up 51 points and 9 comments, ranging from "this isn't half baked, it's basically raw" to a genuinely useful side debate about how AI agents should "see" a UI at all — through screenshots or through the accessibility tree.
Short answer: Deltix is a free, open-beta Mac app where you describe a task in plain English, and an AI agent attempts it in an iOS Simulator, then tells you if a real user could complete it. It's iOS-simulator-only today, has no pricing page yet, and drew mixed reviews on Hacker News — genuinely interesting agent, rough edges in the current build.
I went through Deltix's own site, its beta terms, and the full Hacker News thread rather than just the headline pitch, because "AI agent tests your app" is a crowded claim right now and the gap between the demo and the daily-driver experience is usually where these tools fall apart. Worth saying up front: Deltix's agent runs as a native Mac app tied to Xcode Simulator, so this is built from its documented setup flow and what beta users are reporting, not a multi-week trial run — I don't have a spare Mac with Xcode wired up sitting idle for this piece. When I tested the sign-up flow and re-checked the four onboarding steps against the live site, all of them still matched what's documented below, and the "no credit card, no invite" claim held up too. What follows is what's actually on the page today, what the Show HN crowd flagged, and how it compares to the free alternative most iOS teams could reach for instead.
What you'll need
Deltix's beta only runs on macOS, so you need a Mac with Xcode installed and at least one iOS Simulator you can boot. You'll also need an app you can build and run locally — Deltix tests against a build already sitting on your machine, not a hosted build or a TestFlight link, and the vendor is explicit that "your source and your build stay on your Mac," which matters if your team has any code you can't ship to a third-party cloud. Sign-up is free with no credit card and no invite required, just an email you can verify. If you're doing this for a team rather than solo, budget time to also read Deltix's beta terms, since open-beta software changes fast and without a changelog you'll want to know what you agreed to.
Step-by-step: setting up Deltix's AI-driven testing beta
1. Create a free account
Go to app.deltix.ai and sign up. Deltix's own copy is blunt about the bar here — "Free. No credit card. No invite needed." — you verify your email and you're in.
2. Install the Mac Agent
Deltix runs as a small native Mac app rather than a browser extension or CLI. You install it and sign in with the account you just created; this is the piece that actually drives the simulator, not the web dashboard.
3. Connect a running simulator
Boot any iOS Simulator you already use for development, then point the Mac Agent at it. Deltix doesn't spin up its own simulator instance — it attaches to one you already have running, which keeps the setup close to a normal dev workflow.
4. Write your first task in plain English
The example Deltix gives on its own site is "sign up and send your first message" — a full-sentence instruction, not a script or a selector. The agent takes it from there, tapping and typing its way through the app the way an actual first-time user might.
5. Read the run, not just the pass/fail
Each run captures screenshots as the agent moves through the app, so a failed task isn't just a red X — you can see where the agent got stuck. That matters more than the pass/fail label itself, especially in a beta where the agent's own confusion and a real bug can look identical from the outside.
6. Save a good run as a Playbook
If a run does what you wanted, save it as a Playbook so it replays deterministically later instead of re-improvising the same task from scratch every time. Deltix also has an Experiment mode for running the same task against two builds side by side, which is closer to a regression check than a one-off exploration.
Example prompts you can copy
These follow the same plain-English shape as Deltix's own example task, adapted for a few common first tests:
- "Sign up with a new email and confirm you land on the home screen."
- "Add an item to the cart, go to checkout, and stop before entering payment details."
- "Try to sign in with a wrong password three times and check what the app shows."
- "Send a message, then delete it, and confirm it's gone from the thread."
- "Rotate the device and confirm the layout doesn't break on the settings screen."
The pattern worth copying isn't the exact wording — it's writing the task the way you'd hand it to a new tester, with an explicit stopping point, rather than a vague goal like "test checkout."
Common mistakes to avoid
The most common complaint in the Hacker News thread is worth repeating before you invest setup time: commenter shermozle reported the app opening with the interface greyed out and the listed support email bouncing, a rough first impression for anyone judging the product from that thread alone rather than trying it themselves. Second, don't assume this replaces Android or physical-device testing yet — Deltix is iOS Simulator only today, with physical devices, Android, a CI-friendly CLI, and hybrid app support all listed as planned rather than shipped. Third, watch the cost of the AI agent itself, not just the app subscription — commenter perfectlyFine pointed out that current-generation agents are capable enough for this kind of UI task but can burn through a lot of tokens getting there, so a "free" testing tool can still carry a real bill depending on which model is doing the driving. Fourth, don't treat a passing Task run as a permanent guarantee; save it as a Playbook if you want it to mean the same thing next week, since a one-off run isn't a regression test until you do that. Fifth, remember open-beta software has no changelog or version history published yet, so a workflow that works today can shift under you without notice — read the beta terms rather than assuming stability.
Deltix vs. XcodeBuildMCP at a glance
| Deltix | XcodeBuildMCP | |
|---|---|---|
| Price | Free during open beta, no pricing page published | Free, MIT licensed, open source |
| How you drive it | Its own native Mac Agent + dashboard | MCP server you connect to Claude, Cursor, or any MCP client |
| What it does | AI agent attempts a plain-English task, judges if a user could finish it | Build, install, launch, screenshot, and UI-automate Xcode/iOS projects |
| Simulator support | iOS Simulator only (physical device "next" on the roadmap) | Simulator and physical device deploys today |
| Replay / regression | Playbooks (save a run) and Experiment mode (A/B builds) | No built-in replay layer; you script it yourself |
| Maturity | Open beta, launched on Hacker News August 15, 2026 | Actively developed, v2.0 released, roughly 1,900 GitHub stars |
Both are real, verifiable products as of this writing, not vaporware pitches — I confirmed Deltix's feature list and beta status directly on its own site, and XcodeBuildMCP's license, install path, and version on its project site.
Tools that make this easier
If you're weighing Deltix against a broader set of AI coding and testing tools, my best AI tool for code roundup covers where agent-driven QA fits next to agent-driven coding. For the "screenshot vs. accessibility tree" debate that came up in the Show HN comments, it's the same underlying question I cover in my Cursor vs. Copilot comparison about how much an AI tool actually sees versus infers. Before trusting any vendor's beta claims — including Deltix's — my AI tool reviews piece walks through how I separate a real capability from a demo-only one, and AI tool ratings covers reading community feedback like this Hacker News thread without over- or under-weighting a handful of loud comments. If you want to try an AI agent workflow without committing to Deltix's Mac-only beta yet, free AI tools lists no-cost options worth testing first, and my how to use Claude AI guide is the starting point if you'd rather wire up your own agent, the way commenter alightsoul described doing for airline QA with Claude Desktop and Playwright. If you're already running several AI agents on one codebase, my piece on a local merge queue for parallel Claude Code agents covers the coordination problem that shows up once testing agents and coding agents are both touching the same repo.
My take
Deltix is solving a real problem — most AI-driven testing pitches skip straight to "it writes your test suite," while Deltix's angle of judging whether a real user could finish a task is closer to how a human QA tester actually thinks. But it's an August 2026 open beta with no pricing, no Android, no physical devices, and at least one Hacker News commenter who couldn't get past a greyed-out first screen. That's not disqualifying for a beta — most of them look like this — but it does mean you should treat this as an early look rather than a tool to build a release process around yet. If you're iOS-only, already comfortable with the trust tradeoff of a local agent touching your simulator, and fine with software that has no changelog yet, it's worth the ten minutes to sign up and run one task. If you need Android coverage, physical-device testing, or CI integration today, XcodeBuildMCP or a scripted agent workflow will get you further right now.
Frequently Asked Questions
Is Deltix free?
Yes, for now. Sign-up requires no credit card and no invite, and Deltix hasn't published a pricing page as of this writing — that's typical for an early open beta, but expect that to change once it moves past beta.
Does Deltix work on Android?
Not yet. The open beta covers iOS Simulator only. Android support, physical-device testing, a CI-friendly CLI, and hybrid app support are all listed on Deltix's own roadmap as planned, not shipped.
What's the easiest way to try Deltix?
Sign up, install the Mac Agent, connect a simulator you already have running, and give it one plain-English task like "sign up and send your first message." Don't start with your most complex flow — start with something you can judge as pass or fail at a glance.
Is the Hacker News criticism a dealbreaker?
Not automatically. One commenter hit a greyed-out screen and a bouncing support email, a legitimate red flag for a first impression, but open-beta bug reports are common and don't necessarily reflect the product a week or a month later. Treat it as a reason to test cautiously, not a reason to skip testing entirely.
How is Deltix different from just wiring an AI agent to Xcode myself?
Tools like XcodeBuildMCP give you the raw building blocks — build, install, screenshot, tap — through any MCP-compatible AI client, but you assemble the workflow yourself. Deltix packages that into a dedicated agent with a dashboard, saved Playbooks, and side-by-side Experiment comparisons, at the cost of being a single-purpose, Mac-only, iOS-only tool right now.