Argent's Apache licence covers the source, and the binaries it downloads are proprietary
An agentic toolkit to control, debug, and profile iOS and Android apps. Made by Software Mansion.
At a glance
- What is it?
- Argent is an agentic toolkit that gives an AI assistant direct control of iOS simulators, Android emulators, TVs and Electron apps. Its source is Apache 2.0, but the prebuilt simulator server, a macOS service binary and the iOS dynamic libraries are licensed only for use inside the project, and telemetry is opt-out rather than opt-in.
- Who is it for?
- argent is worth trying if you drive mobile targets from a coding assistant, because the flows feature records a path once and replays it deterministically as a repro or a smoke test, which is the part most agent tooling gets wrong. Two things to settle before you build on it.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- software-mansion/argent3,199 ★ · +247
| Repository | 2026-10-01 | 2026-11-15 |
|---|---|---|
| software-mansion/argent | 3k | 3.2k |
247 Stars in the last 45 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
The Apache licence covers the source and the binaries are separate
The licence section states a mixed model and the split is specific enough to matter. Source code is released under the Apache License 2.0. Proprietary binaries are not: the per-platform simulator server executable, a macOS service binary, and the dynamic library files inside the iOS native devtools package are described as the intellectual property of Software Mansion S.A. and licensed solely for use within this project. Decompiling, reverse-engineering, or redistributing them without explicit written permission is prohibited, and using the toolkit is framed as acknowledging that structure. The detail worth flagging is where the terms live. The repository's licence file is the Apache one. The proprietary carve-out is stated in the readme, so an automated licence scan of the licence file alone reports a clean Apache 2.0 project and misses the restriction entirely.
Telemetry is opt-out and the switch is a subcommand
One line covers the data question, and the direction of the default is the whole content of it. Telemetry is opt-out, and the way to turn it off is a subcommand of the tool itself rather than a configuration file, a build flag, or an environment variable. A privacy notice is linked separately and also lives in the repository as its own file at the root. For a toolkit whose entire purpose is to let an assistant read screens, logs, network traffic and native trees on a device, the collection posture is the first thing to check rather than the last. The mechanism being a subcommand means it is discoverable, and it also means a fresh environment has telemetry on until somebody runs it.
The tool list is absent because the assistant asks the runtime
Here is a design decision worth noticing. The readme does not enumerate what the toolkit can do. Instead of a table of operations, it tells you to ask your assistant what Argent can do, which lists every tool. So the capability surface is discovered at runtime rather than read in advance. The entry point is a single command run through the package runner, which is the whole of the documented install:
npx @swmansion/argent@latest init # pnpm: pnpm dlx @swmansion/argent@latest initFor an agentic tool that is arguably the right call, since a static list in a readme goes stale the moment a version ships, and the readme is describing a zero major version already releasing weekly. The cost is that you cannot evaluate the scope from the repository alone; the readme gives you six capability areas in prose and defers the enumeration to the thing you are trying to evaluate.
Six kinds of target, from a simulator to a Fire TV
The platform reach is wider than the name suggests. The targets are iOS simulators, Android emulators and physical Android devices, televisions covering Apple TV, Android TV and Fire TV, and Electron or Chromium applications. Input adapts to the target rather than assuming a touch screen: a D-pad on television, mouse and keyboard on desktop, and the full set of tap, swipe, gesture, type and hardware buttons elsewhere. Profiling is named per platform stack, covering Hermes, React DevTools, Xcode Instruments, and Perfetto, with the outputs described as renders, CPU hotspots, and hangs. Debugging reaches logs, network traffic including both fetch and native calls, JavaScript evaluation, and the native and React trees.
Two pairs of package scripts are identical
The root manifest holds two pairs of entries that run exactly the same command. The pack script and the pack:mcp script both invoke the same packaging file with the same interpreter. The start script and the start:tool-server script both build one workspace package and then run the development command in another, in the same order. That is four entries doing two jobs. It reads as a convenience rename rather than an oversight, since the colon-suffixed names are the conventional spelling and the bare names are the shortcuts, but it does mean anyone auditing the scripts has two names per action to account for. The workspace glob is similarly explicit about what it leaves out, listing the package directory pattern and then negating two specific packages inside it.
Three scripts fetch binaries and one package is built twice
Getting a local build together means downloading more than writing it. Three separate scripts fetch prebuilt material: one for the simulator server, one for native binaries, and one for the trace processor. Alongside those there are build scripts for the iOS native package and the Android native package. The full local packaging command chains all of it in a fixed order, and the chain contains a detail worth noting: the iOS package is built twice, once with no arguments and once with a flag selecting a TCP transport, so both transport variants of the same package are produced before anything else runs. There is also a git submodule configuration file in the tree, which means this repository uses two different mechanisms for obtaining external code.
knip runs twice with complementary rules and no tolerance
The unused-code check is configured as two invocations rather than one. The first runs with exports and types excluded from the report. The second runs with those included and extended to class members, namespace exports, and namespace types, and with the maximum issue count set to zero. So the pair covers both halves: first find code nothing references, then find exports and type surface nothing consumes, and fail the build outright on the second. Zero tolerance on the stricter pass is the notable choice, since a single stray re-export in a toolkit this size is enough to turn the check red. Linting matches that posture elsewhere, running with warnings treated as errors.
Two release channels and two tests that keep them aligned
The release record shows three tags in under a month: a patch in mid-September, a minor the following week, and another minor in early October, with the last push dated 2026-10-03. So the project is on a weekly minor cadence and still on a zero major version, which together mean you should pin an exact version rather than a range. The script list also names a next distribution tag and includes two tests by name: one for synchronising that tag, and one that concerns the canary version. That is a two-channel setup, a stable line and a canary line, with the tag bookkeeping covered by tests rather than left to a release step.
Editorial conclusion
argent is worth trying if you drive mobile targets from a coding assistant, because the flows feature records a path once and replays it deterministically as a repro or a smoke test, which is the part most agent tooling gets wrong. Two things to settle before you build on it. The licence is mixed, so read the readme rather than the licence file, because the file you would normally audit is the Apache one and the proprietary carve-out lives beside it in prose. And it is a zero major version shipping weekly minor releases, so pin exactly. If telemetry matters to you, note the switch is opt-out and has to be run deliberately.
Frequently asked questions
What is Argent and what does it control?
It is an agentic toolkit that gives an AI assistant direct control of iOS simulators, Android emulators and devices, televisions covering Apple TV, Android TV and Fire TV, and Electron or Chromium apps. It can tap, swipe, gesture, type and press hardware buttons, use a D-pad on television, and use mouse and keyboard on desktop.
How do you install Argent?
One command through the package runner, which creates the project configuration. The documented line runs the init command from the latest version of the published package, with an inline comment giving the equivalent command for users of pnpm. There is also a hosted documentation set covering installation, supported platforms, the command line reference, and editor integration.
What licence is Argent under?
A mixed model. The source code is Apache 2.0, but the per-platform simulator server executable, a macOS service binary, and the dynamic library files in the iOS native devtools package are the intellectual property of Software Mansion and licensed only for use inside the project, with decompiling, reverse-engineering and redistribution prohibited without written permission.
Does Argent collect telemetry?
Yes, unless you turn it off. Telemetry is opt-out rather than opt-in, and the way to disable it is a subcommand of the tool rather than a configuration file or an environment variable. A separate privacy notice is linked from the readme and also exists as its own file in the repository root.
What can Argent actually do, and how do I find out?
The readme deliberately does not enumerate the tools. It tells you to ask your assistant what Argent can do, which lists every tool at runtime. The readme itself names six capability areas in prose: interaction, recorded flows that replay deterministically as a repro or smoke test, OCR and font aware visual regression, profiling across four named stacks, debugging of logs, network and both trees, and React Native build and launch.
Official sources
Where developers are discussing it
Posts on Hacker News, dev.to and Lobsters that link to this repository, found by our scan of those communities. 0 in the last 30 days; newest first.
- This Week In React #301 : TanStack Start, React Aria, Base UI, Next.js | RN Web, Tamagui, One, Skia | Deno, Bun, TS, JPEG XLdev.to · Oct 16, 2026 · 2 points · 0 comments
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/software-mansion-argent)Star history chart for your README
[](https://hysenlabs.com/projects/software-mansion-argent)Was this useful?