In coatless/searcher: Query Search Interfaces and AI Assistants

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "README-"
)

searcher

Overview

The goal of searcher is to provide a search interface directly inside of R. For example, to look up rcpp example numeric vector or ggplot2 fix axis labels call one of the search_*() functions to automatically have a web browser open, go to a search site, and type the query. searcher also provides direct integration with AI assistants through ask_*() functions, allowing you to send queries to ChatGPT, Claude, and other AI services with R-optimized prompts.

By default, the search and ask functions will attempt to search the last error on call if no query is specified.

Installation

The searcher package is available on both CRAN and GitHub. The CRAN version is considered stable while the GitHub version is in a state of development and may break.

You can install the stable version of the searcher package with:

install.packages("searcher")

For the development version, you can opt for:

if(!requireNamespace("remotes")) { install.packages("remotes") }
remotes::install_github("coatless-rpkg/searcher")

Usage

library(searcher)

Search Terms

The search_*() functions can be used to search a query directly from R on major search engines, programming help websites, and code repositories. The following search platforms are supported: Google, Bing, DuckDuckGo, Startpage, Ecosia, rseek, Qwant, Brave, Kagi, X (formerly Twitter), BlueSky, Mastodon, StackOverflow, Posit Community, GitHub, grep.app, and BitBucket. By default, an appropriate suffix for each platform that ensures relevant results to R is appended to all queries. This behavior can be disabled by using rlang = FALSE.

# Searching R project on major search engines
search_google("R project")
search_bing("R project")
search_ecosia("R project")
search_rseek("R project")
search_qwant("R project")
search_brave("R project")
search_kagi("R project")
search_duckduckgo("R project")                           # or search_ddg(...)
search_startpage("R project")                            # or search_sp(...)

# Searching X/Twitter to find out about machine learning for R and in general
search_twitter("machine learning")
search_twitter("machine learning", rlang = FALSE)

# Searching BlueSky to find out about deep learning for R and in general
search_bluesky("deep learning")
search_bluesky("deep learning", rlang = FALSE)

# Searching Mastodon to find out about data visualization for R and in general
search_mastodon("data vis")
search_mastodon("data vis", rlang = FALSE)

# Searching SO for linear regression questions for R and in general
search_stackoverflow("linear regression")
search_stackoverflow("linear regression", rlang = FALSE) # or search_so(...)

# Searching Posit Community for tips
search_posit_community("tips")
search_posit_community("tips", rlang = FALSE)          # or search_posit(...)

# Searching GitHub code for graphs in R and other languages
search_grep("graph")
search_grep("graph", rlang = FALSE)

# Searching GitHub Issues for maps in R and other languages
search_github("maps")
search_github("maps", rlang = FALSE)                     # or search_gh(...)

# Searching BitBucket for assertions in R and other languages
search_bitbucket("assertions")
search_bitbucket("assertions", rlang = FALSE)            # or search_bb(...)

AI Assistants

The package also provides functions to query AI assistants directly from R. The following AI Assistant platforms are supported: OpenAI's ChatGPT, Anthropic's Claude, Perplexity, Microsoft (Bing)'s Copilot, Mistral's le Chat, xAI's Grok, and Meta.ai. These functions open a browser with your query pre-filled, using customizable prompts that help the AI give more effective responses for R programming:

# Get coding help from AI assistants
ask_chatgpt("How to create a ggplot scatterplot with regression line?")
ask_claude("Explain what purrr::map_df does")
ask_perplexity("Compare dplyr vs data.table performance")
ask_mistral("How to handle missing data in R?")
ask_bing_copilot("Write a function to calculate the median")
ask_grok("What is better base R or tidyverse for research?")
ask_meta_ai("What are the best R packages for time series analysis?")

# Search with an error message
tryCatch(
  cor(mtcrs),  # Intentional typo
  error = function(e) ask_claude()  # Will search the error message
)

All AI search functions accept an optional prompt parameter that guides how the AI responds:

# Adding specific instructions to the prompt
ask_chatgpt(
  "Fix this code: ggplot(mtcars, aes(x=mpg, y=hp) + geom_point()", 
  prompt = "You are an R debugging expert. Explain what's wrong step by step."
)

See vignette("using-ai-assistants-with-searcher", package = "searcher") for more details on using AI assistants in searches through searcher.

AI Prompt Management

For those who frequently use AI assistants, searcher provides a prompt management system:

# List available prompts
ai_prompt_list()

# Set a system-level prompt for all AI services
ai_prompt("debugging")  # Use a predefined prompt for debugging

# Create custom prompts
ai_prompt_register("my_prompt", "As an R expert analyzing the mtcars dataset...")

# Check active prompt
ai_prompt_active()

# Clear active prompt
ai_prompt_clear()

See vignette("managing-ai-prompts") for more details on the prompt management system.

Comparison with ellmer Package

While both searcher and ellmer provide AI assistance for R users, they take different approaches:

• searcher opens a web browser with your query pre-filled in the AI service's interface
• No API keys needed
• Works with multiple AI services
• Visual interface for complex interactions
• Requires browser access and being logged in
• ellmer uses API connections to interact with models directly within R
• Requires API keys
• Results returned directly to R console/environment
• Better for programmatic/automated use
• Works in non-interactive environments

These packages are complementary rather than competitive:

• Use searcher for interactive exploration, debugging, and research
• Use ellmer for reproducible workflows, batch processing, and production code

See vignette("faq", package = "searcher") for a more detailed comparison.

Search Errors

searcher offers preliminary support for automatically or manually searching errors that are generated in R. For more robust error search support and to also search warning messages, please use the errorist package.

Automatically

Searching the last error automatically is possible by registering a function as R's error handler via either searcher(site="") or one of the search_*() functions. Thus, when an error occurs, this function will automatically be called. This triggers a new browser window to open with the error term listed in verbatim.

# Using the generic search error handler
options(error = searcher("google"))

# Directly specify the search or ask function
options(error = search_google)

options(error = ask_claude)

Manually

Alternatively, these functions can also be used manually so that the default error dispatch is preserved. In the manual case, you will have to explicitly call the search function. After that, a browser window will open with the last error message as the search query on the desired search portal.

# Search the last error message with Google
search_google()

# Ask an AI assistant about the last error
## Switch into debug mode
ai_prompt("debugging")

## Ask Claude about the last error
ask_claude()

Package Customizations

The ability to customize different operations in searcher is possible by setting values in options() within ~/.Rprofile. Presently, the following options are available:

• searcher.launch_delay: Amount of time between launching the web browser from when the command was issued. Default is 0.5 seconds.
• searcher.use_rstudio_viewer: Display search results in the RStudio viewer pane instead of a web browser. Default is FALSE.
• searcher.default_keyword: Suffix keyword to focus search results between either "base" or "tidyverse". Default is "base".
• AI service prompts:
• searcher.chatgpt_prompt: Default prompt for OpenAI's ChatGPT queries.
• searcher.claude_prompt: Default prompt for Anthropic's Claude queries.
• searcher.perplexity_prompt: Default prompt for Perplexity AI queries.
• searcher.mistral_prompt: Default prompt for Mistral AI queries.
• searcher.bing_copilot_prompt: Default prompt for Bing Copilot queries.
• searcher.grok_prompt: Default prompt for xAI's Grok AI queries.
• searcher.meta_ai_prompt: Default prompt for Meta AI queries.

To set one of these options, please create the .Rprofile by typing into R:

file.edit("~/.Rprofile")

From there, add:

.First = function() {
  options(
    searcher.launch_delay       = 0,
    searcher.use_rstudio_viewer = FALSE,
    searcher.default_keyword    = "tidyverse",
    searcher.chatgpt_prompt     = "You are an R programming expert. Please provide concise answers with code examples.",
    ## Additional options.
  )
}

Motivation

The idea for searcher began as a project to automatically search errors and warnings that occurred while working with R after a conversation among Dirk Eddelbuettel, Barry Rowlingson, and myself. However, there was no search interface that allowed querying directly from R outside of the built-in utils::RSiteSearch(), which only queries http://search.r-project.org/, and the sos package, which queries an off-site user premade database. Both of these options were focused solely on querying R documentation made available by packages. Given the nature of errors generally being undocumented, neither of these approaches could be used. Thus, searcher was unintentionally born to provide a means for errorist, which contains a robust way to automatically searching errors and warnings.

Special Thanks

• Dirk Eddelbuettel for starting the discussion on XKCD Comic 1185: Ineffective Sorts.
• Barry Rowlingson for remarks about functionality.

Publicity

On the #rstats-twitter verse (c.f. https://twitter.com/search?q=%23rstats), searcher has been positively received by community members.

R package "searcher" that automatically searches Stackoverflow for error that you just saw in the console. Cool package, especially for those who learn R :) https://github.com/coatless-rpkg/searcher ... #r #rlang #rstats #rstudio

--- Paweł Przytuła March 23th, 2019. ~292 Retweets and 876 likes (Note, URL updated to new repository location.)

Did you know, using "searcher" package, you could automatically to search stackoverflow, google, GitHub and many more sites for errors, packages or topics. #rstats

--- Shakirah Nakalungi (c.f. https://twitter.com/cynthia_kyra) June 29th, 2019, when she was Rotating Curator for the "We are R-Ladies" twitter account (c.f. https://twitter.com/WeAreRLadies/status/1144921174251581440) ~144 Retweets and 544 likes

Please let us know via an issue ticket about how you are using searcher.

License

GPL (>= 2)

coatless/searcher documentation built on June 2, 2025, 10:37 a.m.