Release Notes: Lár v1.4.0

Release Notes: Lár v1.4.0

"The Polish Update"

This release transforms Lár from a powerful engine into a refined product. It introduces a CLI for instant scaffolding, "Low Code" decorators for cleaner code, and crucial developer experience (DX) improvements.

🚀 New Features

1. The Lár CLI (lar)

Stop manually creating files. Scaffold production-ready agents in seconds.

pip install lar-engine
lar new agent my-bot

This generates a complete project structure with pyproject.toml, .env, and agent.py.

2. "Low Code" Nodes (@node)

Define nodes as simple Python functions.

from lar import node

@node(output_key="summary")
def summarize(state):
    return llm.generate(state["text"])

3. Dictionary Access for State

Interact with GraphState naturally.

# Old
val = state.get("key")
state.set("key", val)

# New
val = state["key"]
state["key"] = val

🛠 Improvements & Fixes

• Robust Concurrency: Fixed a crash in BatchNode when using functools.partial or dynamic Callables. Logging is now safe for all object types.
• Identity Resolution: The package remains lar-engine on PyPI, but the CLI ensures your project is set up correctly to import lar without confusion.

🔄 Backward Compatibility

Safe to Upgrade. Lár v1.4.0 is fully compatible with v1.3 code.

• Existing LLMNode / class-based setups work unchanged.
• You can mix "Old" and "New" styles in the same graph.
• See examples/v1_4_showcase.py for a demonstration.

📦 Installation

pip install lar-engine==1.4.0