Blog & Changelog
Stay updated with WeClaw's latest news, tech insights, and best practices
When AI Learns to Look in the Mirror: Self-Reflection Closed-Loop Architecture from Philosophy to Engineering
A four-tool self-healing architecture (perceive - diagnose - fix - restart) enabling AI to audit its own logs, analyze errors, search source code, and safely restart.
Context Management in Practice: When AI Loses Memory in Million-Token Conversations
Deep dive into LLM Agent context management challenges, from compression strategies and disaster recovery to async summarization.
Pitfall Chronicles: 10 Counter-Intuitive Bugs in Context Management
A collection of 10 counter-intuitive bugs from two context management refactoring rounds, each following symptom - root cause - fix - lesson format.
Proactive Companion Engine: How Five Components Decide When and What to Say
From event-driven to asyncio non-blocking scheduling, revealing the decision engine behind AI proactive care and anti-harassment cooldown.
Performance Optimization: 3x Faster Startup and 60% Memory Reduction
Lazy loading, object pools, LRU caching, and full-chain profiling — startup time reduced from 10s to 3s.
Hybrid Communication Architecture: The Triple Fusion of HTTP+SSE+WebSocket
How WeClaw designs a hybrid communication architecture for efficient collaboration between desktop, PWA mobile, and remote services.
Curing AI Amnesia: Cross-Session Experience Knowledge Graph Makes AI Smarter Over Time
Building ExperienceStore for automatic diagnostic experience recording, semantic retrieval of historical solutions, and cross-session knowledge accumulation.
Tool Registration System Evolution: From Manual Mapping to Config-Driven Auto-Discovery
The complete evolution of the tool registration system from hardcoded mappings to tools.json config-driven auto-discovery.
Compression Thresholds: Why AI Never Compresses Under 1M Windows
In-depth analysis of context compression threshold strategies, revealing the engineering decisions behind 1M window models.
Tool Pairing Integrity: The Most Insidious Bug in ReAct Agents
Uncovering the most hidden bug in ReAct loops — tool_call and tool_result pairing failures — and how to fix it.