Skales
A local-first desktop assistant for carrying out multi-step work across your apps.
Per-dimension scores and reasoning
Evidence: README describes permission controls (Safety Mode, path guard, shell guard, per-agent confinement) but these are claims about the current release; the repository source is a v7 snapshot lacking these mechanisms. Thus, these security features are not present in the source, only asserted. Deductions: least privilege, user confirmation, data flow transparency, sensitive data handling, dependency security, external effects, rollback, and source attribution are not evidenced in the source code.
Evidence: README and SECURITY.md consistently describe the repository as a distribution channel with a v7 snapshot, but the source version (7.1.0) differs from the product version (12.7.1), causing inconsistency. Dependencies are minimal (electron-updater, electron-store) with no lockfile or integrity checks, so dependency availability is unverified. Failure messages: the custom-endpoint/test route provides detailed error handling, but other parts lack evidence. Deductions: dependency security not assessed, version inconsistency.
Evidence: README details many use cases (goals, coding, mobile, teams) for a broad audience. Capability boundaries are described (Safety Mode, permissions) but not in source. Trigger precision: /goal, /code commands are explained but exact triggers not specified. Environment fit: multi-platform support claimed, but source is v7 snapshot, far from current. Deductions: capability boundaries and trigger precision lack source evidence.
Evidence: README is well-structured with installation, feature index, license. License is BSL 1.1 but metadata says NOASSERTION, inconsistent. Version in README is 12.7.1 but package.json is 7.1.0, inconsistent. Known limitations are in SECURITY.md (snapshot not maintained). Maintenance responsibility: author Mario Simic mentioned in SECURITY.md, but no clear maintenance plan. Deductions: version inconsistency, license metadata mismatch.
Evidence: README claims powerful features, but source is v7 snapshot, so actual output cannot be verified. Marginal value: product positioning is unique (local-first, multi-platform), but no real usage evidence. Cost-benefit: free trial, but no performance data. Deductions: output usability not verifiable from source, cost-benefit lacks data.
Evidence: Many claims in README (security features, performance) are not reflected in source, so traceability is low. Cross-source corroboration: only README, no other sources. Fact-inference separation: README distinguishes product description from repository notes, but not clearly which are facts. Deductions: claims lack source support, no cross-source verification.
- The source in this repository is a v7 snapshot, vastly different from the current release (12.7.1), and does not represent the actual product.
- License metadata is NOASSERTION but actual is BSL 1.1, inconsistent.
- Dependency security not assessed; no lockfile or integrity checks.
- Security features (path guard, shell guard) are only claimed in README, not present in source.
What does this agent do, and when should you use it?
Skales is a local-first desktop and mobile AI assistant for Windows, macOS, Linux, Android, and iOS. It accepts work through chat, /goal, /code, /codework, and visual Workflows, with access to approved files, browser, calendar, email, and desktop-control capabilities. The desktop product is described as an Electron application with a Next.js 14 and TypeScript interface, storing data in JSON and SQLite under ~/.skales-data; its mobile app can pair with a desktop by QR code or run on its own. Distribution is centered on signed application binaries, releases, and the issue tracker. The repository also states that its checked-in v7 source snapshot is outdated and not built, so this is a product to adopt as a downloaded application rather than a current self-hosted codebase.
A user can enter /goal to create a background, multi-step task that may resume after the app is closed and can run on a schedule. /code binds a local folder to a chat and offers Plan, Code, Edits, and Auto permission modes; it can read a repository map, change files with edit_file, run test_run, render inline git diffs, and—when approved—use git_commit, git_push, or deploy_project. /codework runs a project-scoped sequence of reading files, planning, writing code, testing, and reviewing changes in a three-panel interface. The built-in browser agent navigates sites, fills forms, and extracts Markdown, while Computer Use controls the screen through screenshots, mouse input, keyboard input, and scrolling, with per-action approval in Safety Mode. Skales can also import SKILL.md or .skill.zip packages, use MCP Servers, and manage WordPress pages, posts, media, WooCommerce products, and SEO metadata through its Connector Plugin.
- A solo developer binds a repository to /code, requests a read-only plan first, then approves edits, tests, commits, and a push within the selected mode.
- An operations specialist records a repeatable desktop procedure once and invokes the saved workflow later with a trigger phrase.
- A personal user creates a /goal task for work that should continue in the background or recur on a schedule.
- A user away from their desk pairs Skales Mobile with a home or office desktop to access that desktop’s tool set through the phone.
- A WordPress content manager connects the Skales Connector Plugin and asks for pages, posts, product changes, media uploads, or SEO updates in natural language.
- A privacy-conscious user selects their own OpenAI, Anthropic, Gemini, or other provider credentials, or uses a local Ollama or LM Studio endpoint.
What are this agent's strengths and limitations?
- Offers native Windows, macOS, and Linux installers plus Android and iOS apps, without requiring Docker or a terminal.
- Combines background goals, chat-bound code work, project-level Codework, browser automation, and desktop control in one application.
- Supports 15+ providers, BYOK, Ollama, and other local or OpenAI-compatible endpoints instead of tying use to one cloud model.
- Supports SKILL.md, .skill.zip, MCP Servers, Agent2Agent, and cross-device Teams for extension and collaboration.
- Uses local-first storage under ~/.skales-data and states that API requests go directly to the chosen provider.
- The repository says its source is an outdated v7 snapshot; current functionality is distributed as closed-source binaries, limiting independent auditing and current-source deployment.
- The product can access high-privilege capabilities including files, browsers, calendars, email, shell commands, and screen control; approval modes exist, but access boundaries still need careful configuration.
- Skales IQ is a trial rather than a permanent documented free model service; ongoing cloud use will generally require a provider account, subscription, or API key.
- Offline operation depends on a local endpoint such as Ollama or LM Studio, so capability and quality depend on the available local model and hardware.
- Licensing evidence conflicts: repository metadata says NOASSERTION, while the README describes BSL 1.1 terms that restrict commercial SaaS and competing products.
How do you install or deploy this agent?
Install the distributed application from skales.app: download the Windows EXE, macOS DMG, or Linux .deb/AppImage; on macOS, drag the application from the DMG to Applications. Install Skales Mobile through Google Play or the App Store. On first launch, Skales IQ can be used as a no-API-key trial; afterwards, configure a provider key in Settings → AI Providers or connect Ollama, LM Studio, KoboldCpp, vLLM/text-generation-webui, or an OpenAI-compatible endpoint. The repository provides no supported commands for building or deploying the current product and explicitly says its v7 source is no longer maintained, built, or used by the shipped app.
How do you use this agent?
Enter /goal followed by a task description to create a background task, for example /goal build me a trading bot. For local project work, use /code <task>, select a folder, then choose Plan, Code, Edits, or Auto; use /codework for a project-scoped coding session. Select Skales IQ, configure BYOK credentials, or connect a local model in Settings → AI Providers. Import a skill by pasting a GitHub URL, selecting a local folder, or pasting SKILL.md content in Agent Skills. Pair the mobile app by scanning the desktop QR code, or use its standalone mobile tools without pairing.
How does this agent compare with similar options?
Skales positions itself against desktop agents that require Docker, a terminal, or a Python CLI, and it includes migration importers for ChatGPT, Claude, Copilot, Gemini, OpenClaw, and Hermes. Its stated differentiators are a native local application, offline-model options, cross-device pairing, and graphical code and workflow interfaces; these comparisons are product positioning claims.