DragonClaw

Possibly the best desktop client for OpenClaw — built with Electron, Vue 3 and Vite.

🌐 Official site: http://www.dragonclaw.cc

English · 简体中文 · 繁體中文 · 日本語 · Русский · 한국어 · العربية · Deutsch

---

📑 Contents

• 🐉 About DragonClaw
• 🔍 Feature Deep-Dive
• 🏗️ Architecture
• ✅ Requirements
• 🚀 Quick Start
• 🛠️ Scripts
• 📦 Build & Release
• 📁 Project Layout
• 🧪 Testing
• 🗺️ Roadmap
• 🤝 Contributing
• 📄 License
• 🌟 Star History
• 🙏 Acknowledgments

---

🐉 About DragonClaw

DragonClaw is a purpose-built desktop client for OpenClaw, developed on top of OpenClaw's official WebSocket protocol API. It is built with a modern Electron + Vue 3 + Arco Design stack and ships with the complete runtime required by OpenClaw — install once, run instantly. If you already have OpenClaw installed locally, DragonClaw detects and reuses it automatically, with zero migration cost.

Beyond being a graphical shell for the Gateway, DragonClaw introduces a number of innovative, product-grade capabilities:

• 🧠 Agents — list, create, update and delete agents; read & write per-agent workspace files.
• 🧩 Skills — browse recommended skills, search the full catalog, install / uninstall with a single click.
• 💬 Multi-conversation — run many sessions in parallel with smart grouping, pinning, unread markers and full history.
• 📁 Session-level Workspaces — set the working directory for the current session directly inside the conversation panel; every artifact OpenClaw produces stays inside that workspace and never pollutes any other space.
• 🌳 Sub-tasks — visualize sub-tasks and sub-sessions spawned by the main agent in real time.
• 📚 Model Configuration — built-in support for all major providers: OpenAI, Anthropic, DeepSeek, Grok, OpenRouter, Groq, Moonshot, Alibaba, Ollama, ModelBus, custom and more.
• 🏪 Agent Store — browse the online catalog, pick the right Agent and install it together with the skills it depends on, in one click.
• 📋 Dashboard — a task board to organize work across agents.
• 📜 Logs — tail gateway logs directly from the app.
• ⚙️ Unified settings — basics, gateway, agents, sessions, tools and security in a single pane.
• 🔄 Update detection — automatically checks for updates to DragonClaw, OpenClaw, Node.js and more at startup.
• 🌐 Remote access — connect to an OpenClaw Gateway on your LAN or on a remote server, authenticated via Token or password.

📖 For full feature walkthroughs (Conversations, Workspaces, Sub-tasks, Model Configuration, Agent Store, Update Detection, Remote Access) see the 🔍 Feature Deep-Dive section below.

---

🔍 Feature Deep-Dive

Below is a closer look at DragonClaw's most distinctive capabilities and the use cases they unlock.

💬 Conversations (Workspaces & Sub-tasks)

The conversation panel is the heart of DragonClaw. It is not just a chat window — it is a full multi-session collaboration system:

• Parallel sessions — open multiple conversations at once, each bound to a different agent or task. Switch from the left sidebar with zero cross-talk.
• Smart grouping — historical sessions are automatically organized into Today, Yesterday, This Week, Earlier — clean and predictable.
• Pin & unread — important sessions can be pinned; unread sessions show a red dot so nothing slips through.
• Streaming output — model replies stream in real time, with reasoning traces and tool-call results rendered inline.
• Historical scrollback — scroll to the top to lazy-load older messages; nothing is ever truncated.
• Message actions — copy, quote, delete and "compact" sessions are all available locally.
• Thinking level — pick a reasoning intensity (high / medium / low) from the top bar to balance latency against quality.
• Sub-task visualization — sub-tasks and sub-agents triggered by the main agent appear as compact cards inside the related message group, with the responsible agent and live status at a glance.

📁 Session-level Workspaces

Workspaces in DragonClaw are set at the conversation level and never pollute any other space. Each session can carry its own working directory, configured directly from the conversation panel — every file OpenClaw reads, writes or produces inside that session stays inside that workspace, with no leakage into other directories or global locations.

A workspace is bound to the current session, not to the agent globally — so two conversations with the same agent can use completely different folders:

• Set from the conversation panel — click the workspace button in the input bar to pick a folder; it is attached to the active session immediately.
• Per-session scope — the path is stored against the session key (projectSpace), so switching sessions shows each conversation's own workspace, not a global one.
• Context that follows — every file read / write and shell command the agent runs inside that session uses the workspace as its root, so the agent never has to ask "where should I save?".
• Open or rebind — the current workspace is always visible in the input bar; click to open it in the system file manager, or rebind to a new directory at any time.
• Empty-state guard — when no workspace is set, the input bar highlights the gap to prevent mis-operations.

🌳 Sub-tasks

When the main agent spawns sub-tasks or sub-agents, DragonClaw tracks and presents them in two complementary views:

• Inline message view — sub-task cards sit directly beneath the originating message, with the executing agent, task title and live status.
• Right-side summary — open the right panel to see every sub-task in the current session and its completion state at once.
• Real-time sync — status transitions (pending → done) refresh live as the Gateway pushes events.

---

🧠 Model Configuration

Model Configuration is another area where DragonClaw shines. Where vanilla OpenClaw forces you to hand-edit config.json, DragonClaw delivers a fully graphical model management experience — and ships with built-in support for all major model providers out of the box:

• Built-in major providers — OpenAI, Anthropic, DeepSeek, Grok, OpenRouter, Groq, Moonshot, Alibaba, Ollama, ModelBus, plus a fully customizable "Custom" provider — all preconfigured, no manual wiring required.
• Unified multi-Provider management — manage dozens of vendors in one place; switch the active model for any session in seconds.
• Provider grouping — entries are auto-grouped by Provider with logo, model count and context length in plain view.
• Rich model metadata — pulled from the models.dev catalog: context window, supported capabilities (tools, vision, …), pricing tier and more.
• Visual CRUD — adding a model is a few fields (ID, display name, max tokens…) — no JSON editing required.
• Per-Provider credentials — API keys and base URLs are isolated per Provider so they never bleed into each other.
• Local models first-class — Ollama is preconfigured with http://127.0.0.1:11434/v1; local LLMs work out of the box.
• Per-session switching — change the primary model for a conversation from its top bar, no Gateway restart needed.

---

🏪 Agent Store (Agents Store)

Browse Agents online and install them together with the skills they depend on — no more hand-cloning templates from GitHub.

DragonClaw ships with a built-in Agent Store that lets you browse and install agents — along with the skills they need — directly from the application:

• Online catalog — explore a curated catalog of official and community agents, all browsable from inside the app.
• Find the right fit — filter by scenario (development, writing, office, ops, …) and pick the agent that best matches your task, rather than settling for whatever happens to be installed locally.
• One-click install for Agents & skills — choose a target agent and it gets installed together with every skill it depends on, so you never have to chase down dependencies by hand.
• Localized customization — installation is fully visual: avatar, ID, working directory, persona description, etc.
• "My agents" view — a dedicated tab lists everything you have created, ready to edit, delete or inspect.
• Online updates — when an installed agent publishes a new version, upgrade with a single click.

---

🔄 Update Detection

DragonClaw bundles a one-stop component updater that automatically runs at startup, sparing you the chore of comparing version numbers across ecosystems:

• Automatic check at startup — every time DragonClaw launches, it pulls the latest versions of DragonClaw, OpenClaw, Node.js and more in a single request.
• Signed source — goes through config.json's api.versionCheck endpoint with the sign=dragonclaw identifier.
• Visible prompts — when a new version is available, the sidebar and settings page show a clear notice with a link to the changelog.
• In-app download & install — download, install and replace entirely from inside the app, without blocking the main window.
• Rollback-safe — failed downloads or signature mismatches never corrupt the existing installation, so you never get stuck on a blank screen.

---

🌐 Remote Access

Whether your OpenClaw Gateway lives on another machine on your LAN or on a remote server, internal device or home NAS, DragonClaw can connect to it securely using a Token or password:

• One-click switch — pick "Remote Mode" from the menu or flip it in the sidebar; you're connected to the target Gateway in seconds.
• Flexible authentication — both Token-based and password-based authentication are supported, matching different Gateway security policies.
• Live status indicator — the sidebar shows Connecting / Connected / Disconnected in real time, with a manual reconnect button.
• Identical local & remote UX — once connected, every feature (agents, sessions, skills, settings, logs…) behaves exactly as in local mode.
• Connection test — before switching, you can probe the target host to avoid wasting time on a misconfiguration.
• Persistent configuration — remote address, port and Token are stored encrypted locally and restored automatically on next launch.

💡 If your remote device has no public IP, pair DragonClaw with Tailscale, ZeroTier, frp or any other tunneling / NAT-traversal solution.

---

📋 Dashboard

When a task needs to be split across multiple agents working in parallel, DragonClaw's Dashboard surfaces every work item on a single task board:

• Multi-column flow — one column per agent, work pushes forward inside its column and cross-agent collaboration is immediately visible.
• Real-time refresh — sub-tasks and sub-agents spawned by the main agent are written back to the corresponding card instantly.
• Status visualization — idle, running, error and done are shown with distinct markers so bottlenecks stand out.
• One-click refresh — the Refresh button in the header pulls the latest state on demand without waiting for a push.

---

🧩 Skill Management

OpenClaw's skill ecosystem is huge — DragonClaw wraps it in a fully graphical Skill Center so you never have to hand-edit configurations:

• Recommended / All / Managed tabs — start from the curated recommendations or browse the full catalog when you know what you want.
• Keyword search — the search box filters across names and multilingual tags.
• One-click install / uninstall — every skill card has an Install / Uninstall button; rollback and reinstall are equally lightweight.
• Author & last-updated metadata — the list shows the author, install count and last update so you can judge maintenance activity at a glance.
• Local cache sync — installed skills are cached locally so the next launch doesn't redownload anything.

---

🖥️ Computer Assistant

Beyond collaborating with the OpenClaw Gateway, DragonClaw also ships a Computer Assistant module that brings AI capability to your local machine:

• Disk cleanup — intelligently scans for large files, caches and reclaimable items, and estimates how much space you can free up.
• Computer tidy-up — lets AI get a complete picture of your machine to help with file categorization and cleanup.
• Software management — a unified view of every installed app, sortable by size, last update or usage, with one-click uninstall for apps you no longer need.
• **Status refr