ClawX Desktop
A no-terminal desktop workspace for configuring, chatting with, and automating OpenClaw AI agents.
Per-dimension scores and reasoning
Evidence shows: README claims secure keychain storage, but no implementation details; update notifications require user confirmation, but not all external actions; data flow partially transparent (e.g., ACP bridge), but not fully disclosed; sensitive data handling mentioned (e.g., key storage), but not detailed; dependency security has pnpm overrides and onlyBuiltDependencies, but no vulnerability scan evidence; external effects (e.g., auto-update, proxy settings) mentioned, but user confirmation not explicit; rollback mechanism not mentioned; source attribution has author and license, but publisher unverified. Deductions: lack of concrete implementation evidence, some claims unsupported by code.
Evidence shows: README and code structure consistent, but no runtime verification; dependency versions pinned, but no availability guarantee; failure messages mentioned (e.g., error handling), but no specific examples. Deductions: static review cannot verify runtime behavior, failure message details insufficient.
Evidence shows: README describes multiple use cases (personal assistant, monitoring, etc.), but no specific configuration examples; capability boundaries described (e.g., disabling web_search), but not comprehensive; trigger precision mentioned (e.g., cron), but not detailed; environment fit has system requirements, but not all platform details. Deductions: scenario descriptions generic, trigger mechanism details insufficient.
Evidence shows: README structure clear, installation instructions present, but missing FAQ and known limitations; naming stable but no version history; license MIT, but no changelog; maintenance responsibility has community links, but not explicit. Deductions: missing FAQ, known limitations, and changelog.
Evidence shows: output usability described (e.g., Markdown rendering), but no actual output examples; marginal value described (e.g., desktop interface), but not compared to CLI; cost-benefit not mentioned. Deductions: lack of actual output examples and cost analysis.
Evidence shows: README claims partially supported by code (e.g., feature list), but no test results; cross-source verification insufficient, relying on single source; facts and inferences not clearly separated. Deductions: lack of test evidence and external verification.
- Publisher identity unverified, proceed with caution.
- Dependency security lacks vulnerability scan evidence, recommend checking dependencies for vulnerabilities.
- Sensitive data handling details insufficient, verify key storage implementation.
- Missing FAQ and known limitations, users may encounter unexpected issues.
- Static review cannot verify runtime behavior, recommend actual testing.
What does this agent do, and when should you use it?
ClawX is a cross-platform Electron 40+ and React 19 desktop application that provides a graphical interface for the OpenClaw AI runtime. It brings chat, provider and model settings, skills, messaging channels, scheduled jobs, workspaces, and diagnostics into one local desktop application. Electron Main owns the ACP stdio chat bridge, Gateway supervision, secret storage, and system integration, while the React renderer accesses host services through typed IPC. It targets macOS, Windows, and Linux, and includes the OpenClaw runtime rather than requiring a separate installation.
On first launch, the Setup Wizard selects a language, adds AI providers, and verifies the configuration; OpenAI can use an API key or browser OAuth for a Codex subscription. Chat communicates with OpenClaw through an ACP stdio bridge owned by Electron Main and renders Markdown replies, attachments, file activity from write, edit, and apply_patch calls, and session-level Changes records. On the Channels page, users configure multiple accounts and per-account agent bindings; on the Cron page, they create recurring or one-time tasks with selectable external delivery targets. The Skills page scans managed, workspace, and extra skill directories and enables or disables skills; the app can also preview authorized local HTML, DOCX, and PPTX files, while larger files and legacy Office formats open through the operating system.
- A personal user wants an OpenClaw assistant for questions, document summaries, or email drafts without configuring the runtime in a terminal.
- A team needs to configure several accounts for messaging channels on one desktop machine and bind each account to a specific agent.
- An operations user needs daily, weekday, or one-time agent jobs created in the Cron page and delivered to supported channel targets.
- A developer wants to inspect OpenClaw file activity, change summaries, and local attachment previews from the desktop chat experience.
- A user needs to manage OpenAI, Anthropic, Z.AI/GLM, or custom compatible providers in one GUI while storing credentials in the system keychain.
What are this agent's strengths and limitations?
- Bundles the OpenClaw runtime with the desktop app and covers setup, Gateway lifecycle, and visual configuration, avoiding terminal use for routine operation.
- Uses an ACP stdio bridge owned by the main process, allowing unfinished responses to keep streaming while the user changes conversations or pages.
- Supports multi-account Channels with per-account agent bindings and Cron jobs with recurring, one-time, and external-delivery options.
- Uses local-first skill management across several OpenClaw skill locations and deploys PDF, XLSX, DOCX, and PPTX processing skills by default.
- Its core runtime is OpenClaw, so it is not an independently adoptable general-purpose agent runtime.
- Source development requires a supported Node.js version, pnpm, and an Electron environment; Linux additionally needs specified system libraries.
- Inline Office previews have documented limits: DOCX pagination may differ from Microsoft Word, PPTX lacks animations, transitions, and media playback, and files over 20 MB are not previewed.
- HTML Preview deliberately blocks links, forms, script navigation, redirects, downloads, network requests, and device permissions, so it is not a general web browser.
How do you install or deploy this agent?
Download a pre-built macOS, Windows, or Linux release from GitHub Releases. To run from source, use Node.js 22.22.3+, 24.15.0+, or 25.9.0+ (Node 24 LTS recommended), plus pnpm 9+ or npm: git clone https://github.com/ValueCell-ai/ClawX.git, cd ClawX, pnpm run init, then pnpm dev. At first launch, use the Setup Wizard to add provider credentials—for example an OpenAI API key or supported Codex-subscription OAuth sign-in—and complete verification before entering the app.
How do you use this agent?
Complete language and provider verification in the Setup Wizard, then create or select a Chat session and send messages to an OpenClaw agent. Use @agent in the main composer to route to another agent, or insert /skill-name. Manage models and credentials in Settings → AI Providers, enable discovered local skills in Skills, add channel accounts and bindings in Channels, and create Recurring or Once jobs in Cron. For connectivity through a proxy, configure Settings → Gateway → Proxy; in Developer Mode, run openclaw doctor --json from Settings → Advanced → Developer.
How does this agent compare with similar options?
Compared with using OpenClaw through its command-line workflow, ClawX provides a desktop GUI for providers, skills, channels, scheduled jobs, and chat; OpenClaw remains the underlying agent runtime.