foundryR

A tidy, API-first R interface to Microsoft Azure AI Foundry. Build AI-powered applications with chat completions, the newer Responses API, structured extraction, web-grounded answers, embeddings, content safety, and image generation - all returning tibbles that integrate seamlessly with tidyverse and tidymodels workflows.

Features

• Chat completions - Interact with GPT, Claude, Llama, Mistral, DeepSeek, Cohere, and other models
• Responses API - Use Microsoft Foundry's v1 /openai/v1/responses endpoint for stateful turns, tools, and structured outputs
• Structured extraction - Convert free text into schema-constrained tidy columns for annotation and research workflows
• Web-grounded responses - Use the Responses API web_search tool and get citations as tidy list-columns
• Text embeddings - Generate vector embeddings for semantic search, clustering, and ML
• Content safety - Moderate content, detect hallucinations (groundedness), and protect against prompt injection
• Image generation - Create images with DALL-E models
• tidymodels integration - Use step_foundry_embed() to add embeddings to your ML pipelines

Installation

Install the development version from GitHub:

# install.packages("pak")
pak::pak("farach/foundryR")

Quick Start

Configure credentials

library(foundryR)

# Set credentials for current session
foundry_set_endpoint("https://your-resource.openai.azure.com")
foundry_set_key("your-api-key")

# Verify setup
foundry_check_setup()

For persistent configuration, add to your .Renviron file:

AZURE_FOUNDRY_ENDPOINT=https://your-resource.openai.azure.com
AZURE_FOUNDRY_KEY=your-api-key

Chat with a model

foundry_chat("What is the tidyverse?", model = "gpt-4o-mini")
#> # A tibble: 1 x 7
#>   role      content                          model finish_reason prompt_tokens
#>   <chr>     <chr>                            <chr> <chr>                 <int>
#> 1 assistant The tidyverse is a collection... gpt-4 stop                     10
#> # i 2 more variables: completion_tokens <int>, total_tokens <int>

Use the Responses API

The newer Microsoft Foundry Responses API supports stateful turns, built-in tools, and schema-constrained output through the v1 endpoint:

first <- foundry_response(
  "Define catastrophic forgetting.",
  model = "gpt-4.1"
)

foundry_response(
  "Explain it for a college freshman.",
  model = "gpt-4.1",
  previous_response_id = first$response_id
)

Extract structured data

Use JSON Schema to turn free text into analyzable variables:

schema <- list(
  type = "object",
  properties = list(
    sentiment = list(type = "string", enum = c("positive", "negative", "neutral")),
    entities = list(type = "array", items = list(type = "string"))
  ),
  required = c("sentiment", "entities"),
  additionalProperties = FALSE
)

foundry_extract(
  c("I love using R with Azure.", "The workflow was slow and confusing."),
  schema = schema,
  model = "gpt-4.1"
)

Search the web with citations

web_answer <- foundry_web_search(
  "What changed recently in Azure AI Foundry Responses API?",
  model = "gpt-4.1"
)

web_answer$citations[[1]]

foundry_web_search() uses Grounding with Bing services. Microsoft documents that this can leave compliance/geographic boundaries and incur additional costs.

Generate embeddings

texts <- c("I love R programming", "R is great for statistics")
foundry_embed(texts, model = "text-embedding-3-small")
#> # A tibble: 2 x 3
#>   text                       embedding      n_dims
#>   <chr>                      <list>          <int>
#> 1 I love R programming       <dbl [1,536]>    1536
#> 2 R is great for statistics  <dbl [1,536]>    1536

Compute similarity

embeddings <- foundry_embed(texts, model = "text-embedding-3-small")
foundry_similarity(embeddings)
#> # A tibble: 1 x 3
#>   text_1               text_2                     similarity
#>   <chr>                <chr>                           <dbl>
#> 1 I love R programming R is great for statistics       0.912

Content Safety

foundryR integrates with Azure AI Content Safety for responsible AI features:

# Content moderation
foundry_moderate("Sample text to analyze")

# Hallucination detection
foundry_groundedness(
  text = "AI-generated response",
  grounding_sources = "Source document",
  query = "User question",
  task = "QnA"
)

# Prompt injection detection
foundry_shield(user_prompt = "User input to check")

Image Generation

Create images with DALL-E:

result <- foundry_image(
  "A serene mountain landscape at sunset",
  model = "dall-e-3"
)
foundry_save_image(result, "landscape.png")

tidymodels Integration

Add text embeddings to your ML pipelines:

library(tidymodels)

recipe(sentiment ~ text, data = reviews) |>
  step_foundry_embed(text, model = "text-embedding-3-small") |>
  step_normalize(all_numeric_predictors())

Learn More

• Getting Started - Setup and first API calls
• Responses API - Stateful turns, structured extraction, and web search
• Text Embeddings - Semantic search and similarity
• Content Safety - Responsible AI features
• Image Generation - Creating images with DALL-E
• tidymodels Integration - ML pipelines with embeddings
• Function Reference - Complete API documentation

License

MIT