Core Features

QuantClaw combines powerful AI capabilities with local-first execution and robust error handling.

🧠 Intelligent Conversation

Multi-turn dialogue with full context awareness:

• Context Management: Automatic compaction and pruning to stay within token limits
• Thinking Mode: Enable extended reasoning for complex problems
• Session History: Persistent conversation tracking and replay
• Memory Integration: Automatic context retrieval from knowledge bases

quantclaw run "Help me analyze this code"

💾 Persistent Memory System

Advanced memory management inspired by human cognition:

Memory Types

• User Memory: Persistent user profiles and preferences
• Agent Memory: Long-term learned patterns and behaviors
• Working Memory: Current session context
• Workspace Files: SOUL, IDENTITY, HEARTBEAT, MEMORY, SKILL

Memory Operations

quantclaw memory search "user preferences"
quantclaw memory get workspace/MEMORY.md

Automatic Context Pruning

• BM25-based relevance scoring
• Overflow compaction with 3-attempt retry
• Budget-based context management
• Thinking level auto-adjustment

🌐 Browser Control

Full Chrome DevTools Protocol integration:

Browser tools (browser.launch, browser.execute, browser.screenshot) are built-in agent tools invoked automatically during a session. They are not direct CLI commands.

Capabilities

• Page Navigation: Load and control URLs
• DOM Interaction: Click, type, submit forms
• JavaScript Execution: Run custom scripts
• Screenshot Capture: Visual verification
• Cookie Management: Session persistence
• Network Monitoring: Intercept requests

💻 System Integration

Execute commands and manage system resources:

Command Execution

The bash and process tools are built-in agent tools (not direct CLI commands). The agent uses them automatically when executing tasks.

File Operations

# Workspace files are accessed via the agent or directly from ~/.quantclaw/agents/main/workspace/
ls ~/.quantclaw/agents/main/workspace/
cat ~/.quantclaw/agents/main/workspace/MEMORY.md

Environment Access

quantclaw config get llm.model

🔌 Plugin Ecosystem

Extensible architecture with Node.js sidecar:

Built-in Plugins

• Bash Tools: Safe command execution
• Browser Control: CDP integration
• Web Search: Multi-provider search (Tavily, DuckDuckGo, Perplexity)
• Web Fetch: HTML scraping with SSRF protection
• Memory Search: BM25-based knowledge retrieval
• Cron Scheduling: Automated task execution
• Process Management: Background job control

Custom Plugin Development

quantclaw skills list
quantclaw skills install ./my-plugin

🔄 Multi-Provider LLM Support

OpenAI-compatible and Anthropic APIs with automatic failover:

Supported Providers

• OpenAI (and compatible endpoints): qwen-max, GPT-4, DeepSeek, local Ollama, etc.
• Anthropic: Claude Haiku, Sonnet, Opus
• Ollama / Gemini: Registered (stub, not fully implemented)

Any OpenAI-compatible API endpoint can be used by setting providers.openai.baseUrl.

Provider Selection

{
  "llm": {
    "model": "anthropic/claude-sonnet-4-6",
    "maxIterations": 15,
    "maxTokens": 4096
  },
  "providers": {
    "anthropic": { "apiKey": "sk-ant-..." },
    "openai": { "apiKey": "sk-..." }
  }
}

Automatic Failover

• Exponential backoff on failures
• Profile rotation and health monitoring
• Fallback chain execution
• Token usage accumulation

🛡️ Enterprise Security

Production-grade security controls:

Role-Based Access Control (RBAC)

RBAC is enforced at the gateway level. Roles include:

• agent.admin, operator.read, operator.write, viewer

Tool Permissions

• Per-tool allow/deny rules
• Command pattern matching
• Execution approvals
• Audit logging

Sandboxing

• Process isolation
• File system restrictions
• Network policies
• Resource limits

Multi-level Authorization

• Agent-level: agent.admin, agent.operator, agent.viewer
• Tool-level: tool.{name}.read, tool.{name}.write
• Session-level: User-scoped operations

📊 Usage Tracking

Comprehensive monitoring and analytics:

quantclaw usage cost          # Total token costs
quantclaw sessions usage      # Session-by-session breakdown
quantclaw sessions list              # List sessions
quantclaw logs tail          # Real-time logs

Metrics Collected

• Token usage by provider and model
• Request latency
• Error rates
• Provider health
• Session duration

📱 Multi-Channel Integration

Connect across communication platforms:

Supported Channels

• Discord: External subprocess adapter
• Telegram: External subprocess adapter
• Web dashboard: Built-in at port 18801

Channel Configuration

{
  "channels": {
    "discord": {
      "token": "YOUR_BOT_TOKEN"
    },
    "telegram": {
      "token": "YOUR_BOT_TOKEN"
    }
  }
}

🎯 Workflow Automation

Scheduled and triggered automation:

Cron Jobs

quantclaw cron add "daily-report" "0 9 * * *" "Send daily summary"
quantclaw cron list
quantclaw cron remove TASK_ID

Hooks and Callbacks

• Pre/post-processing hooks
• Custom event handlers
• Lifecycle callbacks
• Error recovery handlers

📊 Performance Optimizations

• C++17 Native: Direct compilation for target platform
• Efficient Memory: Context compaction and pruning
• Fast Startup: Lazy initialization of modules
• Concurrent Operations: Multi-threaded request handling
• Smart Batching: Grouped API calls

---

Next: Learn about the architecture or start building plugins.