Create an ellmer tool for secure R code execution — securer_as_ellmer

Wraps a SecureSession as an ellmer::tool() definition so an LLM can execute R code in a sandboxed environment. The tool accepts a single code argument (a string of R code) and returns the result.

Usage

securer_as_ellmer_tool(
  session = NULL,
  tools = list(),
  sandbox = TRUE,
  limits = NULL,
  timeout = 30
)

Arguments

session: A SecureSession object. If NULL (the default), a new session is created with the given tools, sandbox, and limits arguments. When you supply your own session, those arguments are ignored.
tools: A list of securer_tool() objects to register in the session. Only used when session is NULL.
sandbox: Logical, whether to enable OS-level sandboxing. Only used when session is NULL. Defaults to TRUE.
limits: Optional named list of resource limits. Only used when session is NULL.
timeout: Timeout in seconds for each code execution, or NULL for no timeout. Defaults to 30.

Value

An ellmer ToolDef object that can be registered with chat$register_tool().

Examples

if (FALSE) { # \dontrun{
library(ellmer)

# Basic usage: LLM can execute R code in a sandbox
chat <- chat_openai()
chat$register_tool(securer_as_ellmer_tool())
chat$chat("What is the mean of the numbers 1 through 100?")

# With custom tools available inside the sandbox
tools <- list(
  securer_tool("fetch_data", "Fetch a dataset by name",
    fn = function(name) get(name, "package:datasets"),
    args = list(name = "character"))
)
chat$register_tool(securer_as_ellmer_tool(tools = tools))
chat$chat("Load the mtcars dataset and compute the mean mpg.")

# With a pre-existing session
session <- SecureSession$new(sandbox = TRUE)
chat$register_tool(securer_as_ellmer_tool(session = session))
# ... use chat ...
session$close()
} # }