Architecture
A dedicated architecture to secure and improve OpenClaw for production and business environments.
Explore the technical foundations of our autonomous AI deployment. Each domain is isolated, independently managed, and documented in detail.
Memory Pipeline
6-stage processing pipeline with 4 parallel search layers. From raw sessions to curated knowledge. Includes automatic extract-memories and multi-level compaction.
Explore MemoryInter-Agent Bridge
Governor daemon, rate-limiting, webhook protocol. How Max and Eva communicate safely. Multi-agent orchestration via the coordinator pattern.
Explore BridgeAgents
Two instances, 7 sub-agents, scheduled crons, iMessage one-shot architecture. Markdown agent definitions and adversarial verification agent.
Explore AgentsSecurity Deep Dive
Exec-guardian, locked files, PIN, mail-reader sandbox, imsg-watcher, scope isolation. Permission classifier and PreToolUse hooks.
Explore SecurityMonitoring Deep Dive
Dashboard v2, Bridge Analytics, behavioral scoring, drift detection, radar visualization. Automatic report simplification.
Explore MonitoringEasyClaw v2 brings 13 major features to the autonomous agent framework. These capabilities are integrated into every domain of the architecture: memory, security, agents and monitoring.
Memory
NEWAutomatic memory extraction, multi-level context compaction, semantic selection and autonomous dreaming.
extract-memories, compaction, memory-selection, auto-dream
Security
NEWIntelligent permission classifier, blocking PreToolUse hooks, and adversarial verification agent.
permission-classifier, hooks, verification-agent
Agents
NEWMarkdown definitions, coordinator orchestration, automatic retry/failover and task system.
agent-definitions, coordinator, retry-failover, task-system
Quality
NEWImplementation verification and automatic report simplification.
verify, simplify
An autonomous AI agent running 24/7 without constant supervision raises a fundamental question: who controls what, and how? Our answer is a strict three-level hierarchy where authority flows exclusively top-down. No component can escalate its own privileges.
Human Supervisor
Absolute control. Can stop, reconfigure or replace any component. Authority cannot be bypassed by the system.
Technical Supervisor
Intermediate governance layer. Defines rules, limits and agent permissions. Continuous oversight.
Executing Agent
Operates within the defined framework. No autonomy outside granted permissions. Each session is isolated.
We publish tools built in production under an open-source license. These tools are extracted directly from our deployment -- they run every day on our infrastructure before being shared.
Exec Guardian
Runtime shell command enforcement daemon. Three policy tiers, fail-closed, OS-level immutability.
Live Dashboard
Single-file web server for real-time monitoring. Bridge logs, security decisions, memory status.
Bridge Logging
Inter-agent communication scripts with full JSONL audit trail. Webhook transport, log rotation.