QuantClaw

English

中文

English

中文

Appearance

QuantClawHigh-Performance AI Agent Framework

C++17 native implementation of OpenClaw with persistent memory, browser control, and plugin ecosystem

Get Started
View on GitHub
QuantClaw
⚡

High Performance

C++17 native binary with minimal overhead — no interpreter, no GC pauses, tiny memory footprint

🧠

Context Management

Automatic compaction, BM25 memory search, budget-based pruning, and overflow compaction retry

🌐

Browser Control

Real Chrome DevTools Protocol over WebSocket — navigation, JS execution, screenshots, and more

🔌

Plugin Ecosystem

Full OpenClaw plugin compatibility via Node.js sidecar — tools, hooks, services, HTTP routes, gateway methods

🛡️

Sandboxed Execution

RBAC authorization, setrlimit sandbox, exec approval, path allowlist/denylist, and rate limiting

📱

Multi-Platform

Runs natively on Ubuntu and Windows, with Docker Compose support for production deployments

🔄

Multi-Provider LLM

OpenAI and Anthropic fully supported; provider/model prefix routing; exponential backoff failover

💬

Channel Adapters

Connect Discord and Telegram bots via external subprocess adapters using the WebSocket RPC protocol

🎯

OpenClaw Compatible

Works with OpenClaw workspace files, skill format, JSONL sessions, and the WebSocket RPC protocol

Why QuantClaw? ​

QuantClaw is a C++17 reimplementation of the OpenClaw AI agent ecosystem — built for performance and low memory footprint while staying compatible with OpenClaw workspace files, skills, and the RPC protocol.

  • Native Performance: C++17 binary, no Node.js runtime overhead
  • OpenClaw Compatible: Workspace files, SKILL.md format, JSONL sessions, and WebSocket RPC
  • Plugin Ecosystem: Run OpenClaw TypeScript plugins via Node.js sidecar (TCP loopback IPC)
  • Production Ready: Daemon mode, Docker support, RBAC, sandbox, failover

Quick Comparison ​

FeatureQuantClawOpenClaw
LanguageC++17TypeScript/Node.js
Runtime OverheadMinimalNode.js VM
Memory FootprintLowMedium
Plugin Support✅ Sidecar (TCP IPC)✅ Native (in-process)
CLI CompatibilityCore commandsFull
Workspace Files✅ All 8✅ All 8
Config FormatJSON5 + ${VAR}JSON5 + ${VAR} + $include

Key Capabilities ​

Intelligent Conversation ​

  • Multi-turn dialogue with automatic context management
  • Dynamic iterations (32–160) based on task complexity
  • Context compaction with 3-attempt overflow retry
  • Session persistence in JSONL format

Persistent Memory ​

  • 8 workspace files: SOUL.md, MEMORY.md, SKILL.md, IDENTITY.md, HEARTBEAT.md, USER.md, AGENTS.md, TOOLS.md
  • BM25 memory search across all workspace files
  • Budget-based context pruning

Browser Automation ​

  • Real Chrome DevTools Protocol (WebSocket)
  • Page navigation, DOM interaction, JS execution, screenshots
  • Graceful fallback when no browser is available

System Integration ​

  • bash tool with sandboxed execution
  • apply_patch for code patching (*** Begin Patch format)
  • process for background job management
  • web_search (Tavily → DuckDuckGo cascade) and web_fetch

Plugin Ecosystem ​

  • 24 lifecycle hook types (void/modifying/sync)
  • Custom tools, services, providers, commands, HTTP routes, gateway methods
  • TCP loopback IPC — works identically on Linux, macOS, and Windows

Getting Started ​

bash
# Clone and build
git clone https://github.com/QuantClaw/QuantClaw.git
cd QuantClaw
./scripts/build.sh

# Initialize
./build/quantclaw onboard --quick

# Install the background service and chat
./build/quantclaw gateway install
./build/quantclaw gateway start
./build/quantclaw agent "Hello, introduce yourself"

For detailed instructions, see Getting Started.

Community & Support ​

License ​

QuantClaw is released under the Apache 2.0 License.

Built with ❤️ in C++17. Inspired by OpenClaw.

Released under the Apache 2.0 License.

Copyright © 2024-2026 QuantClaw Contributors