LLMR: A Unified Interface for Large Language Models in R

LLMR is a unified interface package for large language models specifically designed for R, and it has been published on CRAN, supporting installation via simple commands. It addresses the pain points of R users accessing LLMs—without switching to a Python environment or writing tedious HTTP code, users can seamlessly use multi-provider models like GPT, Claude, and Gemini in their familiar R environment. Core features include unified interface calls, structured output, embedding vector retrieval, and session management, helping data scientists improve analysis efficiency.

In the field of data science and statistical analysis, R holds an important position, but LLM tools and SDKs prioritize Python support, making the R ecosystem relatively lagging. R users face a dilemma: either switch to a Python environment or write tedious HTTP call code to access model APIs, and the fragmented workflow seriously affects analysis efficiency.

Multi-provider Support

LLMR adopts a unified interface design. Whether using OpenAI's GPT, Anthropic's Claude, or Google's Gemini, all are called via the same function, without needing to care about underlying API differences. Once configured, you can freely switch models.

Structured Output Support

Native JSON Schema support allows models to return data in a predefined format (e.g., a structure containing "tags" and "confidence"), eliminating the hassle of subsequent parsing.

Embedding Vector Function

Supports embedding model providers like Voyage, and provides batch processing capabilities to efficiently handle large-scale text data, suitable for tasks such as text similarity calculation and semantic search.

Conversation History and Session Management

The built-in chat_session object maintains multi-turn conversation context and automatically manages message history, making it easy to build interactive assistants or automated reporting tools.

Automated Data Annotation

Use the structured output function to batch classify and annotate text data (e.g., sentiment analysis of customer reviews), and the standardized JSON results can be directly integrated into data frames for subsequent statistics.

Intelligent Report Generation

Combine R's statistical capabilities with LLM's text generation capabilities: R handles data processing and chart generation, while LLMR converts results into natural language descriptions, enabling seamless collaboration.

Semantic Search Enhancement

Add semantic search capabilities to traditional data frames via the embedding vector function: convert text fields into vectors to achieve similarity matching based on meaning, going beyond keyword matching.

Robust Error Handling

Built-in exponential backoff retry mechanism: automatically waits and retries when encountering API rate limits, avoiding interruptions to the analysis process due to occasional network issues.

Parallel Processing Capability

Configure multi-worker parallelism via setup_llm_parallel to significantly improve the processing efficiency of batch model calls.

Type-Safe Response Handling

API responses are encapsulated as llmr_response objects, providing convenient methods like as.character() and tokens() to safely extract information, avoiding the hassle and error-proneness of directly handling raw JSON.

LLMR has a gentle learning curve. Its API design follows R language conventions, with intuitive function names and comprehensive documentation, allowing R-savvy data scientists to get started quickly. It seamlessly integrates with the tidyverse ecosystem: data frames can be directly used as input, and outputs can be easily converted to tibble format. In the future, as R gains popularity in fields like bioinformatics and financial analysis, LLMR is expected to become a key infrastructure for intelligent upgrades in these areas.

LLMR is created and maintained by open-source community developer asanaei, using the MIT license. The code is open-source and community contributions are welcome. It fills the gap in the R language's LLM toolchain, allowing R users to enjoy productivity improvements brought by LLMs without giving up their familiar toolchain, reducing technical migration costs while accelerating the innovative implementation of AI-enhanced analysis processes.