Inside QClaw and EasyClaw: How These AI Agent “Shrimps” Are Built

QClaw (Tencent)

Download the macOS installer from https://claw.guanjia.qq.com/. The first screen shows a loading animation and the UI is identified as Electron‑based. After scanning a QR code for login the app requests an internal‑test invitation code. By modifying the packaged code (changing the function that validates the invite code to always return success) and repackaging, the invitation check can be bypassed. A community version that removes the check is available at https://github.com/haroldneo/OpenQClaw and a Skills package that automates the bypass is at https://github.com/PeanutSplash/qclaw-skip-invite.

When QClaw runs on a machine that already has OpenClaw installed, it detects the OpenClaw process, offers to associate with it, and can read or modify OpenClaw’s configuration files. Existing conversation data from OpenClaw appears in QClaw’s UI, and messages sent from QClaw are logged in the OpenClaw backend console, confirming tight integration.

QClaw ships three built‑in Skills: environment check, service command documentation, and startup rules. The qclaw‑rules skill implements rule‑based service management and is highlighted as a design reference.

Deployment: connects to an existing OpenClaw instance or installs one automatically.

Isolation: association may alter the existing OpenClaw configuration files.

UI resources: stored locally.

Upgrade: upgrade OpenClaw for engine features; upgrade QClaw for client features.

EasyClaw (Leopard)

Download the domestic version from https://easyclaw.cn/. EasyClaw also uses Electron, but the UI is loaded from a remote HTML page identified in the source code. The front‑end appears to be built with Vue, Element UI components, and the micro‑frontend framework wujie.

EasyClaw embeds a Node.js runtime and launches an OpenClaw process on port 10089, avoiding conflicts with other OpenClaw instances. The configuration file .easyclaw/.easyclaw.json contains the token required to access the OpenClaw console at http://localhost:10089/chat?token=xxxxx, allowing direct inspection of the backend service.

The built‑in Skill Store is fully localized in Chinese. English identifiers for skills can be found in the .easyclaw/skills directory.

Deployment: bundles OpenClaw, Node, and dependencies; runs as an independent instance.

Isolation: does not affect any existing OpenClaw configuration; uses its own port 10089.

UI: remote HTML loaded at runtime, not stored locally.

Upgrade: client must be updated when a new OpenClaw version is released.

Comparison

Both agents share the core OpenClaw engine but differ in integration depth and user experience. QClaw acts as a configurable WeChat‑focused client that can associate with an existing OpenClaw installation, potentially modifying its configuration. EasyClaw provides a plug‑and‑play experience with an embedded OpenClaw instance, a Chinese‑localized skill marketplace, and remote UI loading.

Analysis of installation directories, UI resources, and skill definitions reveals the implementation choices for each product and offers concrete clues for developers who wish to fork, extend, or build their own agents.