Skip to main content

YingClaw v2.0 Released: New Workflow Engine and Multi-Model Routing

YingClaw v2.0 is officially released. This version represents a major leap forward in task execution quality and controllability.


🆕 Key Updates

1. Workflow Engine

v2.0 introduces a mandatory four-stage execution pipeline:

Requirements Clarification → Project Assessment → Plan Confirmation → Parallel Execution

Each stage has clear deliverables, and every step is traceable. No more "the Agent went off the rails and you have no idea what happened."

2. Multi-Model Routing

You can now assign different models to different scenarios:

model_routing:
default: gpt-4o
code_review: claude-opus-4
translation: deepseek-chat
simple_task: llama-3-8b # Local model to save costs

High-frequency simple tasks use cheaper models; complex reasoning uses top-tier models — balancing performance and cost.

3. Memory System Refactored

Upgraded from a single MEMORY.md file to a structured memory database:

  • core: Permanent core memories
  • user: User preference memories
  • daily: Session notes with 7-day auto-expiry

Query efficiency improved 10x, with semantic search support.

4. Security Policy Upgrades

New command allowlist/blocklist mechanism with granular permission control by directory and command type.


📊 Performance Comparison

Metricv1.xv2.0
Complex task completion72%91%
Avg. execution steps12.37.8
Memory usage480MB320MB
Startup time3.2s1.1s

📦 How to Upgrade

npm update -g yingclaw
yingclaw --version # Should output v2.0.x

Configuration files are backward-compatible. No additional changes needed.


What's Next

v2.1 roadmap: visual workflow editor, skill marketplace official launch, native WeCom integration. Stay tuned!