Less duplicate code. Smaller bundles. 📦
We've improved Rspack's chunking algorithm to reduce duplicate modules across chunks.
In a large production project, this change resulted in: • -2.53% less raw JS size • -1.51% fewer chunks
One skill. Five platforms.
Midscene Skills gives Claude Code & Codex eyes + hands across browser, desktop, Android, iOS and HarmonyOS.
Same chat-driven prompt. No MCP setup.
🌲Rspack's tree shaking just got smarter 🦀
Two recent optimizations remove more unused code:
1. Inlined booleans can make branch-local dependencies inactive
2. Namespace default re-exports now preserve property-level usage
More precise bundles, less dead code
Everyone is building AI agents.
Very few are building the infrastructure needed to actually run and train them at scale.
Today the veRL team at ByteDance open-sourced Uni-Agent: A unified framework for building, running, and training real-world agents.
Why this matters:
Most agent frameworks focus on orchestration.
Uni-Agent focuses on infrastructure.
It unifies:
- agent execution
- environment interaction
- large-scale concurrency
- RL training
into one reusable stack.
That means the same system can scale from:
prototype → deployment → RL optimization
without rebuilding the stack every time.
The project is still early, but the direction is extremely promising for:
- coding agents
- browser agents
- computer-use agents
- large-scale agent training
Open sourced by the team behind veRL (21k+⭐).
Repo: https://t.co/VM2ah2dImZ
Would love feedback, contributors, and early adopters.
#OpenSource #AI #AIAgents #RL #LLM #ByteDance
🚀 Lance is released!
Lance is a native unified multimodal model from ByteDance Intelligent Creation Lab, supporting image/video understanding, generation, and editing in a single framework.
With 3B activated parameters, Lance explores multi-task synergy for unified multimodal modeling, covering X2T, X2I, and X2V tasks with strong performance across generation, editing, and understanding benchmarks.
Check it out and , give a star:
Homepage: https://t.co/67ONkLIwXl
arXiv: https://t.co/yhRXjh92ME
HuggingFace: https://t.co/oAEPPhpK9y
GitHub: https://t.co/KiZNTkI58e
#Lance #MultimodalAI #UnifiedMultimodalModel #ImageGeneration #VideoGeneration #OpenSource
Two years building Midscene.js, we made a bet we should've made earlier:
UI automation has to move from "reading the DOM" to "looking at the screen."
So in our 1.0 release (Dec 2025), we dropped DOM compatibility entirely.
Rspack 2.0 has landed 🦀
The growth since 1.0 is wild:
📈 100K → 5M weekly downloads
⚡️ ~100% faster builds
📉 devServer: 192 deps → 1
And new stuff hits hard:
🧩 Pure ESM core, modern ESM library output
⚛️ Experimental RSC support
...
Lean, Fast, Modern Web Bundling 🚀