# ClawPanel: a single-binary control panel for OpenClaw agents and their chat channels

> ClawPanel is a Go program that embeds a React admin interface, talks to a local SQLite database and exposes a Gin API with WebSocket push, so you can watch and drive OpenClaw agents, plugins, skills and more than twenty chat channels from a browser. It no longer bundles the OpenClaw runtime: the agent and its channel gateways are external processes the panel installs, supervises or merely observes.

**zhaoxinyi02/ClawPanel** — 🐾 ClawPanel — OpenClaw AI 助手可视化管理面板。Go 单二进制部署，支持 20+ 通道统一管理，跨平台，实时日志监控。

- Repository: https://github.com/zhaoxinyi02/ClawPanel
- Website: https://clawpanel.cc
- Stars: 854 · Forks: 139
- Language: Go
- License: NOASSERTION
- Published: 2026-10-20 · Updated: 2026-10-20 · Language: en
- Canonical page: https://hysenlabs.com/projects/zhaoxinyi02-clawpanel

## What ClawPanel is, and the shift away from bundling the runtime

ClawPanel is a visual management panel for the OpenClaw AI assistant, described as a single standard edition focused on a lightweight panel taking over and on managing external runtimes. The stack line is compact: a Go single binary, React 18, TailwindCSS, SQLite, WebSocket real-time push, cross-platform.

The single most consequential fact in the README is the move away from bundling. ClawPanel now only publishes the standard edition, the panel stays lightweight, and the OpenClaw runtime is no longer packaged into the installer. OpenClaw, NapCat, plugins and channel runtimes are all managed as external resources: the panel can guide their installation, and it can take over an existing gateway.

Installation, update and release assets all go through GitHub Releases, and the scripts no longer depend on the old self-built mirror or third-party sync sources.

The Lite distribution line has been discontinued. ClawPanel from now on publishes only the standard version, no longer bundles the OpenClaw runtime, and no longer provides the Lite build, install, update or packaging scripts. Historical Lite versions remain in the older releases and changelogs for users who installed them, for reference and migration.

The rationale is stated in the architecture section: ClawPanel no longer packages the OpenClaw runtime. The panel is responsible for installation guidance, configuration management, state observation and process orchestration, while OpenClaw and the related channels remain external runtimes, which makes them easier to upgrade, migrate and troubleshoot. That is a sensible separation for a project that has to keep up with a fast-moving agent runtime, and it also means the panel can attach to a gateway somebody else is running.

## The architecture: one binary, seven subsystems

The architecture diagram in the README is the clearest description of what the program does, and it reads as a single binary sitting between a browser and a set of external runtimes.

Inside the binary there are several distinct subsystems. The React management interface is embedded with go:embed, so the whole thing ships as one file with no separate asset deployment. The Gin API, WebSocket and JWT login handle the transport and the session. SQLite stores events, tasks, conversations and panel data.

The Process Manager covers the state of and control over OpenClaw, NapCat and the gateway. The Config Manager handles reading and writing openclaw.json, backups, and schema compatibility cleanup. There is a Plugin, Skill and ClawHub subsystem for plugins, skills, the marketplace and local directory management. Workflow, AI Company and Panel Chat cover multi-agent collaboration and task orchestration. And an Updater handles visual updates of both ClawPanel and OpenClaw.

Below the binary sit three external things: the external OpenClaw with its gateway, agents, skills, plugins, memory and dreaming; the channel runtimes, including QQ via NapCat, WeChat, Feishu, WeCom and Telegram; and an optional Hermes for platform, session, profile, task and log management.

The dependency list in go.mod confirms the stack precisely. The direct requirements are gin for the HTTP API, golang-jwt for the login tokens, gorilla/websocket for the real-time push, and modernc.org/sqlite, which is the pure-Go SQLite implementation that lets the binary be built with CGO disabled. The module path is github.com/zhaoxinyi02/ClawPanel and it declares Go 1.24.0.

The Makefile shows how that turns into a release. The frontend is built with Vite into cmd/clawpanel/frontend/dist, the Go build runs with CGO_ENABLED=0 and -trimpath, and the version and edition are stamped into the binary through ldflags pointing at the internal buildinfo package, with a separate set of flags for Windows.

## Channel management and the plugins around them

The channel page is described as unified management for more than twenty channels: QQ, WeChat, Feishu, DingTalk, WeCom, Telegram, Discord, WhatsApp, Slack, Matrix, LINE, Twitch and others.

QQ gets the most specific treatment. It supports the NapCat Docker and Windows Shell run modes, and provides login state, a QR code, OneBot configuration, token validation and reconnection handling. That is the channel most users of this kind of tool actually want, because QQ is where the personal-account use case sits, and it is also the channel with the most moving parts.

The plugin-style channels, Feishu, WeCom and the official QQ bot, support version switching, install detection, save validation and a prompt to restart the gateway.

The Hermes console covers platform configuration, health checks, sessions, tasks, logs, profiles and action management, and the README describes it as suitable for gradual migration or for running two stacks side by side during a trial.

The AI configuration side is broad. Model provider management supports OpenAI, Anthropic, Google Gemini, DeepSeek, Volcengine Ark, Tongyi Qianwen, OpenRouter and Ollama. The provider cards handle quick add, rename, headers, API type, model lists and compatibility fields.

Identity documents are edited online, covering files such as IDENTITY.md and USER.md, which is what adapts a multi-agent workspace. Browser control offers two safety presets, a managed openclaw profile and disabling browser control, and the README gives the reason plainly: to reduce the risk of mistakenly taking over the system browser.

One detail in the config manager is worth isolating, because it is the kind of thing that breaks integrations silently. Before writing openclaw.json, the panel automatically cleans up legacy fields the new schema does not accept, while preserving key configuration such as models, Dreaming and the browser.

## Multi-agent routing, the monitor topology and the workflow centre

The Agent workbench covers creating, editing and deleting agents, setting a default, editing core files, viewing skill, channel and Cron snapshots, and looking at recent conversations.

Binding routing rules support a structured form and an advanced JSON mode, and can be enabled, disabled, reordered and located when a rule has an error. The routing previewer takes context fields such as channel, sender, peer, guild, team, account and roles, and shows which agent would be hit. That is the feature that makes a multi-agent setup debuggable: without a preview you are guessing which binding a message will match.

The Monitor topology diagram visualises the relationships between channels and agents, agent to agent links, and the default fallback, with zoom, drag and detail inspection.

The Workflow Center is the task orchestration layer. It supports template management, AI-generated templates, a visual canvas, event flows, step details and control over running instances. The nodes cover input, wait_user, approval, ai_plan, ai_task, analyze, summary, publish, image_generate and skill, which is a useful indication of what the workflow model is for: it is a supervised pipeline with human approval steps in it, not a free-form agent loop.

Complex tasks can take over the original conversation, with confirmation, progress write-back, pause, resume, retry, approval, continue, and file or image artefact return. The AI Company layer sits on top with a company overview, team management, and task list and detail views, using a multi-role approach to break down, execute, review and write back tasks.

Panel Chat covers single-agent conversation, multi-agent group chat, knowledge context binding, shared context and conversation persistence. Long task status is pushed over WebSocket in real time, and the activity log aggregates OpenClaw events, channel events, workflow events, system events and bot replies, with filtering, search and SQLite persistence.

## Installation, the default password and the first-run questions

The quick start is one command on Linux and macOS, fetched with curl and run with sudo:

```bash
curl -fsSL https://raw.githubusercontent.com/zhaoxinyi02/ClawPanel/main/scripts/install.sh -o install.sh
sudo bash install.sh
```

The automatic sequence is described as download the binary, install to /opt/clawpanel, register the system service, enable start on boot, configure the firewall and start. On Windows the equivalent is a PowerShell command, and the note says the script is compatible with PowerShell 5.1 and above, which is the version Windows ships with, and that it fetches the latest version from GitHub automatically with no need to specify a version number.

The FAQ table in the README is the part to read before installing, because it lists the questions everyone asks. Starting the service with systemctl needs sudo, and it asks for the Linux system password rather than the panel password. The default panel login password is clawpanel, and the advice is to change it after the first login. A blank panel or a connection failure means checking the service state, opening port 19527 in the firewall, and checking the cloud security group. On macOS, the cannot verify developer error is cleared by running sudo xattr -d com.apple.quarantine /opt/clawpanel/clawpanel.

For uninstalling, the answer is sudo systemctl stop clawpanel followed by sudo rm -rf /opt/clawpanel.

The AI assistant in the bottom right of the panel is worth mentioning because the README recommends using it first: it has a built-in FAQ knowledge base and is meant to help with installation, configuration, channel and runtime problems.

The full FAQ lives at docs/FAQ.md, and the API documentation at docs/API.md. There is also an English version of the README at README_EN.md, which matters given that the primary one is in Chinese.

## Limits: the licence, the disclaimer and the state of the project

The limits here are more than usual and the README is direct about them.

The licence is the first thing to check. The badge in the README header says CC BY-NC-SA 4.0, which is a Creative Commons licence with non-commercial and share-alike conditions, while GitHub shows the licence as a custom one it cannot classify, with the text in the LICENSE file. The project is also described as a Pro edition with the release tags named pro-v5.5.2, pro-v5.5.1 and pro-v5.5.0, and the Makefile stamps an edition of pro into the binary.

The disclaimer is more important than the licence, and it is in a callout at the top of the README. The project is for learning and research purposes only, and commercial use is strictly prohibited. Using third-party clients to sign in to QQ or WeChat may violate those services' agreements and carries a risk of account bans, so the advice is to test with a secondary account. The author states that no reverse engineering was performed, that the project is an integration of open source projects, and that no responsibility is accepted for any consequences. Downloading and using it means agreeing to the full disclaimer.

The second limit is the maturity of the release line. The tags are v5.5.x with the newest at 5.5.2 from 2026-06-08, and the last push to the repository was on 2026-06-08. A version number above five suggests a long history, but the visible release names are all Pro, and the Lite line has been retired, so the naming has been through at least one restructuring.

The third is the platform reach. The binary targets Linux, macOS and Windows with systemd, launchd and Windows Service, but the installation scripts are written for Linux and macOS and for Windows PowerShell, and the FAQ notes that macOS needs a quarantine attribute cleared by hand. There is no mention of a container image in the quick start, although a docker/ directory exists in the tree.

The fourth is the password. The default panel password is clawpanel and it ships in the documentation, so any deployment that does not change it is effectively unauthenticated to anyone who has read the FAQ.

## ClawPanel against running OpenClaw directly and against a general-purpose admin panel

Two comparisons make the choice clearer.

The first is against running OpenClaw without a panel. The README describes the panel's job as installation guidance, configuration management, state observation and process orchestration, with the runtime staying external. If you run one agent on one channel and are comfortable editing openclaw.json, the panel is overhead: it is a Go service to deploy, a web server to secure, a default password to change and a database to back up, in exchange for a nicer view of a configuration file you could read.

What the panel buys you is multiplicity. Twenty channels, several agents, routing rules that need to be understood, a workflow canvas, plugin and skill management and a message centre are a lot of state to hold in your head or in a text editor. And the ability to observe without restarting, with the activity log aggregating events across OpenClaw, channels, workflows, the system and bot replies, is the thing that makes a multi-agent setup debuggable.

The second comparison is against a general-purpose admin or observability panel. The difference in approach is domain specificity. A generic panel gives you logs, metrics and a process list. ClawPanel gives you the things that only make sense for this system: the routing previewer that tells you which agent a given message would hit, the schema cleanup that keeps openclaw.json valid across OpenClaw versions, the workflow node palette, and the config checker that scans OpenClaw, NapCat, models, tokens, ports, WebSocket and HTTP settings and offers a one-click fix.

The cost of that specificity is the obvious one. It only works for OpenClaw, it only knows about the plugins and channels the maintainers have wired in, and it inherits the upstream's own version churn. The v5.5.1 release note is instructive here: the focus was fixing compatibility with the new OpenClaw configuration schema, protecting the Dreaming mode configuration and completing the GitHub-only install and update path after the Lite line was retired. That is the shape of the maintenance burden, and it is the number to weigh before depending on the panel.

## Conclusion

Adopt ClawPanel if you are running OpenClaw across several chat channels and want one browser view of agents, plugins, skills and logs, with the panel able to supervise the runtime rather than merely observing it. Do not adopt it if you need a single-agent setup you can manage by hand. Verify first that the channels you use are among the twenty listed, that your platform is one of Linux, macOS or Windows with a supported service manager, and that you read the disclaimer, since the project states it is for learning and research only and that using third-party clients to sign in to QQ or WeChat may breach those services' terms.

## FAQ

### How do I install ClawPanel on Linux or macOS?

Fetch the install script with curl and run it with sudo. It downloads the binary, installs it to /opt/clawpanel, registers a system service, enables start on boot, configures the firewall and starts it. On Windows the equivalent is the PowerShell one-liner, which needs PowerShell 5.1 or above and fetches the latest release automatically.

### Does ClawPanel still bundle the OpenClaw runtime?

No. ClawPanel now publishes only the standard edition and no longer packages the OpenClaw runtime into the installer. OpenClaw, NapCat, plugins and channel runtimes are managed as external resources that the panel can guide through installation or take over, and the Lite distribution line that bundled it has been retired.

### What licence and usage restrictions apply to ClawPanel?

The README header badge shows CC BY-NC-SA 4.0 while GitHub shows the licence as a custom one it cannot classify, with the text in the LICENSE file. The README also carries a disclaimer that the project is for learning and research purposes only, that commercial use is strictly prohibited, and that signing in to QQ or WeChat through third-party clients may breach those services' agreements and risk account bans.

## Sources

- [Issues](https://github.com/zhaoxinyi02/ClawPanel/issues)
- [Project website](https://clawpanel.cc)
- [README](https://github.com/zhaoxinyi02/ClawPanel/blob/main/README.md)
- [Releases](https://github.com/zhaoxinyi02/ClawPanel/releases)
- [zhaoxinyi02/ClawPanel on GitHub](https://github.com/zhaoxinyi02/ClawPanel)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/zhaoxinyi02-clawpanel