Glonce

Deltix launches AI agent that tests mobile apps from plain English tasks

toolsThe Glonce editorsHacker News🔥53Glonce rating53Impact45Novelty50Relevance85Surprise30Players20Credibility70

Deltix is a testing tool, launched on Show HN, built around a single mechanic: an AI agent attempting a user task inside a running simulator or device, then reporting whether a real user could have completed it. A developer writes a task in plain English, such as "sign up and send your first message," and the agent tries it while acting on what it sees on screen, the way a person would. The agent runs locally against the developer's iOS Simulator, and Deltix states it never accesses the source code, the build, or the signing identities. Screenshots and the run record are stored in the user's account for review and replay, and can be deleted at any time; users can also bring their own model key to route inference traffic away from Deltix's own billing. Setup is a small native app: the developer signs in with the account created during signup, and the agent then attaches to any iOS Simulator already running on the Mac. Deltix organizes its testing around three modes built on that same agent-attempts-a-task core. In ad-hoc mode, a developer building a new flow writes a task and watches the agent try it, reading a screenshot of wherever it got stuck, without scheduling a separate user research session. Any successful ad-hoc run can be saved as a Playbook and replayed deterministically against every subsequent build from a dashboard, returning a pass or fail result in seconds. A third mode runs the same task against two builds of the app side by side, showing which build a user can complete and which one breaks before a design disagreement escalates to a full production A/B test. Beyond the Mac simulator, Deltix can run tasks on a real iPhone attached to a Mac, on an Android emulator, and can replay Playbooks from GitHub Actions, GitLab, or CircleCI as part of a CI pipeline; React Native and Flutter apps are treated the same as native iOS apps. During the current open beta, tasks run against a simulator on the developer's Mac. The product is free, requires no credit card, and needs no invite; access follows email verification.

Key facts

  • Developers describe a task in plain English and Deltix's AI agent attempts it on an iOS Simulator, reporting whether a real user could complete the task.
  • The agent runs locally against the simulator, and Deltix states it never accesses the developer's source code, build, or signing identities.
  • A successful run can be saved as a Playbook and replayed deterministically against every build from a dashboard, returning a pass or fail result in seconds.
  • Beyond the Mac simulator, Deltix supports a real iPhone attached to a Mac, an Android emulator, and Playbook replay from GitHub Actions, GitLab, or CircleCI, treating React Native and Flutter the same as native iOS.
  • The product is free during an open beta, with no credit card or invite required, and users can bring their own model key or use Deltix's.

Why it matters

Deltix targets a gap between scripted end-to-end tests, which check that code paths execute, and real user research sessions, which are slow to schedule but actually judge whether a task is usable. By having an AI agent attempt a plain-English task the way a person would and report where it gets stuck, Deltix aims to give developers a fast, repeatable signal about usability on every build, not just correctness. The idea is not new in shape (AI-driven UI testing agents have appeared elsewhere), but Deltix packages it specifically around mobile simulators, real-device runs, and CI replay.

Who it affects

Mobile developers and teams building native iOS and Android apps, or cross-platform apps in React Native or Flutter, who currently rely on manual QA passes or scripted UI tests to catch usability problems before shipping a build.

How to use it

Setup is a small native app: create a Deltix account, verify the email, and sign in once. The agent then attaches to any iOS Simulator already running on the developer's Mac, or can be pointed at a real iPhone attached to the Mac or an Android emulator. Playbooks, saved from successful runs, can be replayed from GitHub Actions, GitLab, or CircleCI. The product is free during the open beta, with no credit card and no invite required; developers can use Deltix's own model or supply their own model key to route inference traffic off Deltix's bill.

How solid is it

The only source here is Deltix's own product page, describing its own features in its own words; nothing in the text is independently verified. The Show HN discussion had drawn 24 points and 7 comments at the time of writing, a modest early reception. The page names no company, founders, or team, gives no launch date, version number, or funding information, and cites no customer names, testimonials, or usage figures such as how many developers or runs the product has seen.

Risks and caveats

The privacy claims, that the agent runs locally and never touches source code, builds, or signing identities, are self-reported by Deltix with no independent confirmation available from this page. No pricing beyond "free" during the open beta is stated, so what happens after the beta ends is unknown. The page does not say which AI model powers the agent unless a developer supplies their own key. As with any LLM-driven UI agent, task attempts can in principle behave inconsistently between runs; Playbooks are Deltix's stated answer to that, promising deterministic replay of a run that has already succeeded once.

“The agent runs locally against your simulator. We never access your source, your build, or your signing identities.”

— Deltix product page