Skip to contents

Changelog

Source: NEWS.md

LLMAgentR 0.3.1 (Upcoming Release - February 2026)

New Features

  • Custom Agent Builder API
    Added build_custom_agent() for creating user-defined graph/state-machine agents from custom node functions, static edges, and conditional edges.

  • Exported State-Graph Primitives
    Exported StateGraph(), make_node(), make_edge(), make_command(), and interrupt() to support custom workflow construction.

  • LangGraph-Style Mermaid Export
    Added compile_graph() and as_mermaid() for Mermaid visualization, including optional subgraph grouping and graph direction control.

  • PNG Graph Export Utility
    Added save_mermaid_png() to render Mermaid graph text (or compiled graph objects) into PNG files via Mermaid CLI (mmdc).

  • Reusable Code Agent with Builder Pattern
    The build_code_agent() function now supports a builder pattern. If user_input is omitted, the function returns a reusable agent (closure) that can handle multiple queries interactively. This allows for flexible integration in iterative coding workflows, similar to the existing build_researcher_agent().

  • Reusable Interpreter Agent
    The build_interpreter_agent() function now also supports a builder pattern. When code_output is omitted, it returns a reusable interpreter function that can process multiple outputs (tables, model results, charts, etc.) on demand. This provides a consistent interface for explaining analysis results dynamically.

Improvements

  • Dual Invocation Support
    Users can now call both build_code_agent() and build_interpreter_agent() in two ways:
    1. One-shot mode – Provide the input (e.g., user_input, code_output) for immediate interpretation or code generation.
    2. Builder mode – Omit the input to return a persistent agent function.
  • Enhanced Documentation
    The roxygen examples have been expanded to illustrate both usage patterns clearly for both agents, including structured return objects and typical use cases.

Compatibility