Skales: a closed-source desktop agent shipped through a public repository
Personal AI desktop agent for Windows, macOS, Linux, Android & iOS. Set a goal, it works on its own. Teams (pair two desktops, agents + humans), Agent2Agent, Workflows, Codework, multi-agent orgs, desktop + browser automation. 15+ AI providers, BYOK. No Docker, no terminal. Agent Skills (SKILL.md). Migration importer. Recurring autonomous tasks.
At a glance
- What is it?
- Skales is a personal AI agent for Windows, macOS, Linux, Android and iOS distributed under BSL 1.1. The repository carries releases, the changelog and the issue tracker, while the checked-in TypeScript tree is a frozen v7 snapshot that is not what runs on your machine.
- Who is it for?
- Adopt Skales if you want an agent on a personal machine and you are willing to treat the signed installer as the product, not the repository. Do not adopt it if your procurement or security process requires reading the source that ships, or if you need a licence that permits offering the agent to third parties as a service.
- Can I use it commercially?
- Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
- Is it still maintained?
- Yes. The repository last received commits 5 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What the repository is, and what it is not
The README states this plainly: the repository is where Skales is distributed, carrying releases, the changelog, install guides and the issue tracker. The product is the signed application downloaded from skales.app. The source tree checked in here is described as a historical v7 snapshot that is not maintained, not built, and not what runs on the machine. The current release listed is v12.9.26, dated 2026-09-06, with v12.9.25 and v12.9.21 shortly before it. The gap between v7 and v12 is the whole point: anyone cloning this repository to audit the agent's behaviour is reading a version that stopped being relevant several major releases ago. The GitHub licence field reports NOASSERTION, while the README and the badge both say BSL 1.1. That mismatch is worth noting because automated licence scanners will read the repository metadata, not the prose.
The problem Skales is aimed at: agents that need a terminal to start
The README frames the target user by contrast. Its comparison table lists typical AI agents as requiring Docker, a terminal and a Python CLI, with RAM usage between 1.5GB and 3GB, and Linux or Docker as the host requirement. Skales positions itself as a download of an EXE, DMG or AppImage that you double-click, at roughly 300MB of RAM, native on Windows, macOS and Linux, with a stated time to first task of 30 seconds. The audience is stated explicitly as non-developers: the README says the product is made for everyone from 6 to 60+. That framing explains several design choices, including the absence of a terminal requirement, the built-in trial called Skales IQ that needs no API key, and the migration importer for people arriving from OpenClaw, Hermes Agent or ChatGPT. If you are comfortable with Docker and a CLI, the specific problem Skales solves is not one you have.
How the agent runs: goals, background steps, and where execution happens
The mechanism described in the README is a goal handed to a background loop. The example given is /goal build me a trading bot, which the documentation says runs in the background across many steps and picks up where it left off. That is a resumable task queue rather than a single request-response turn, and it is the core of the autonomous claim. A second mode, /code, binds a folder to any chat, with inline diffs and one-click undo, which is a narrower and more auditable unit of work than a free-running goal. Mobile pairing works by QR code: the README says your phone can drive the desktop's full tool set, or run standalone. Execution is local by default, with the README stating that files never leave the machine, and offering three provider paths: the built-in Skales IQ trial, a bring-your-own-key setup across 15+ providers, or fully offline operation with Ollama. The topics list confirms the local-first and MCP angles, though the README does not document the MCP surface in the excerpt available here.
Installing it: download links, the importer, and what is not documented
Installation is presented as platform downloads rather than a build. The README lists skales.app/download/latest-windows, latest-mac-arm64, latest-mac-intel and latest-linux, plus Google Play and the App Store listings for mobile. There is no package manager command, no npm install, and no build step described, which is consistent with a closed-source product. The one configuration path the README names concretely is the migration importer, reached through Settings > Import from Another Tool, for users switching from OpenClaw, Hermes Agent or ChatGPT. Beyond that, the excerpt available here does not document config file locations, environment variables, or CLI flags. Given the stated no-terminal positioning, that is likely intentional, but it means you cannot provision Skales declaratively or check its settings into version control. For a single desktop that is fine. For a fleet, it is a real gap, and the README does not address it.
The licence is the limitation that matters most
Skales is closed source under BSL 1.1, free for personal use. That combination is coherent for the stated audience but constrains a different one. A source-available licence with a personal-use grant typically restricts commercial or hosted use, and the README does not spell out the commercial terms in the excerpt available here. The practical consequence: an individual can install the signed app and use it, but an organisation that wants to embed, resell or host the agent for others needs to read the actual licence text on skales.app before assuming anything. The repository metadata reporting NOASSERTION makes this worse for anyone doing automated licence compliance, because the machine-readable signal disagrees with the human-readable one. There is also a support-boundary limitation: the README points security reporters to SECURITY.md and says it defines what is in scope, which implies out-of-scope reports will be closed. Combined with a frozen source snapshot, an external researcher cannot independently verify a fix against the code that shipped.
Where a terminal-first agent is the better choice
The comparison in the README is honest about the trade it is making: Skales trades configurability for approachability. An agent you start from a shell, define in a config file, and run inside a container gives you reproducible environments, scriptable setup, and a diffable record of what changed. Skales gives you a double-click installer, a goal loop you cannot inspect, and settings behind a GUI. If your work is CI-driven, if you need the agent to run headless on a server, or if you need to pin and audit an exact version of the code executing on your machine, the terminal-first model is the correct one and Skales is the wrong tool. The 300MB RAM figure in the README is a claim about the desktop app, not a benchmark, and there is no published measurement here to check it against. The same applies to the 30-second time-to-first-task figure: it is a positioning statement, not a result you can reproduce from the repository.
Release cadence and the cost of tracking it
The release history shows v12.9.26 on 2026-09-06, v12.9.25 on 2026-09-03, and v12.9.21 on 2026-09-02. Three releases in five days, with named builds (Grip, Backbone, Cockpit 2.0). That cadence is the maintenance story, and it cuts both ways. You get fixes quickly. You also get a moving target: version numbers at the patch level are changing every couple of days, and the README's own framing of Cockpit 2.0 as a named release suggests features are still being restructured rather than settled. For a personal install this is low cost, since the app updates itself. For anyone who needs to validate a specific build before rolling it out, the cadence means validation work has a short shelf life. The changelog at CHANGELOG.md is the place to check what actually moved between two versions, and it is the only artefact in this repository that reflects the shipping product rather than the v7 snapshot.
Who should install it, and what to check first
Skales fits a person who wants an agent with file, browser, calendar and email access on their own machine, does not want to run Docker, and is satisfied with a signed binary they cannot read. The mobile pairing and the goal loop are the two features that distinguish it from a chat client with plugins. It does not fit a team that needs to audit the executing code, provision agents from configuration, or offer the agent to third parties under a permissive licence. Before installing, read the BSL 1.1 terms on skales.app rather than the badge, and confirm that Settings > Import from Another Tool handles the tool you are migrating from, since the README names only OpenClaw, Hermes Agent and ChatGPT. If your decision depends on reading the source, this repository will not support it: the tree here is v7, and the product is v12.
Editorial conclusion
Adopt Skales if you want an agent on a personal machine and you are willing to treat the signed installer as the product, not the repository. Do not adopt it if your procurement or security process requires reading the source that ships, or if you need a licence that permits offering the agent to third parties as a service. Before installing, verify two things: the current BSL 1.1 terms on skales.app rather than the badge in the README, and whether the migration importer under Settings > Import from Another Tool covers the tool you are leaving. The repository will not answer either question for you, because the code in it is the v7 snapshot.
Community notes