btw.md

client	sonnet haiku opus sonnet-4.5 sonnet-4 opus-4.5 gemma4 qwen3.6 qwen-3-5-35b qwen-3-5-9b glm-4-7-flash nemotron-3-nano-4b lm-glm4v qwen3 gpt-oss-20b provider model api_args aws_bedrock us.anthropic.claude-sonnet-4-6 additionalModelRequestFields thinking type budget_tokens enabled 4000 provider model api_args aws_bedrock us.anthropic.claude-haiku-4-5-20251001-v1:0 additionalModelRequestFields thinking type budget_tokens enabled 4000 provider model api_args aws_bedrock us.anthropic.claude-opus-4-5-20251101-v1:0 additionalModelRequestFields thinking type budget_tokens enabled 4000 provider model aws_bedrock us.anthropic.claude-sonnet-4-5-20250929-v1:0 provider model aws_bedrock us.anthropic.claude-sonnet-4-20250514-v1:0 provider model aws_bedrock us.anthropic.claude-opus-4-5-20251101-v1:0 provider model base_url name openai_compatible google/gemma-4-26b-a4b http://remy.local:1234/v1 LM Studio provider model base_url name preserve_thinking openai_compatible qwen/qwen3.6-35b-a3b http://remy.local:1234/v1 LM Studio true provider model base_url name openai_compatible qwen/qwen3.5-35b-a3b http://127.0.0.1:1234/v1 LM Studio provider model base_url name params openai_compatible qwen/qwen3.5-9b http://127.0.0.1:1234/v1 LM Studio reasoning_effort medium provider model base_url name openai_compatible zai-org/glm-4.7-flash http://127.0.0.1:1234/v1 LM Studio provider model base_url name openai_compatible nvidia/nemotron-3-nano-4b http://127.0.0.1:1234/v1 LM Studio provider model base_url name openai_compatible zai-org/glm-4.6v-flash http://127.0.0.1:1234/v1 LM Studio provider model base_url name openai_compatible qwen/qwen3-vl-30b http://127.0.0.1:1234/v1 LM Studio provider model base_url name openai_compatible openai/gpt-oss-20b http://127.0.0.1:1234/v1 LM Studio
tools	skills files_list files_read files_write files_replace docs_help_page docs_package_help_topics
options	skills subagent paths ~/.agents/skills tools_allowed docs files_search files_list

Overview

btw is an R package that helps humans and LLMs work together with R by providing utilities to describe R objects, package documentation, and workspace state in LLM-friendly plain text. The package offers a flexible collection of tools that can be used interactively (copy-paste workflows), programmatically (direct function calls), or as enhanced chat clients (via ellmer or MCP servers).

The primary goal is creating a collection of tools useful to both LLMs and humans when working together with R, with an emphasis on flexibility of usage across different workflows and platforms.

Quick Reference

• Project type: R Package
• Language: R (≥ 4.1.0)
• Key frameworks: ellmer (LLM chat integration), mcptools (Model Context Protocol), shiny and shinychat (chat app)

Purpose and Design Philosophy

btw prioritizes flexibility of usage through multiple entry points:

• btw() - Interactive copy-paste workflow: gather context from R and paste into any chat interface
• btw_tools() - Register tools with ellmer chat clients for custom applications
• btw_client() / btw_app() - Batteries-included chat clients with your preferred LLM provider, model, and project context
• MCP server - Expose tools to third-party coding agents like Claude Desktop or Continue via btw_mcp_server()

Project configuration via btw.md files provides conversation stability across sessions by defining default provider, model, tools, and project-specific instructions. These files are treated as instructions for coding assistants and help avoid repeating context.

btw also serves as a laboratory for discovering best practices in LLM tool design - output formats and approaches evolve based on experimentation with what works best across different models.