Lár v1.5.0 Release Notes

Lár v1.5.0 Release Notes

(Major Feature Release - February 19, 2026)

"The Fractal Update"

v1.5.0 introduces Fractal Agency and True Parallelism, transforming Lár from a linear workflow engine into a system capable of massive, concurrent, self-organizing complexity.

Major Features

1. Fractal Agency (Recursive Dynamic Nodes)

Agents can now design Sub-Agents at runtime. A DynamicNode can instantiate another DynamicNode (or a whole graph) which executes and returns results up the chain.

• Use Case: A "Manager" agent spawns a "Coder" agent and a "Reviewer" agent on the fly to solve a problem it wasn't pre-programmed for.
• Implementation: DynamicNode now supports recursive instantiation and execution.

2. True Parallelism (BatchNode)

Lár now supports multithreaded execution. The BatchNode spins up independent threads for each branch of your graph.

• Performance: Run 5+ sub-agents simultaneously (e.g., searching web, writing code, checking logs) instead of sequentially.
• Safety: Each thread gets a deep-copied State, preventing race conditions. Merging is handled automatically.

Examples

• NEW: examples/advanced/fractal_polymath.py
  • A comprehensive showcase where a Manager orchestrates a "Cipher Specialist" and a "Poet" in parallel to solve a complex task.

Internal Changes

• Tests Reorganized: The tests/ directory is now structured into unit, integration, and scenarios for better maintainability.
• Core Patch: node.py and dynamic.py updated to support _pending_concurrent_ids and recursive execution loops.

Lár v1.5.0 represents a paradigm shift. Your agents are no longer just chains; they are living organisms that can grow.