Generate a plot from a natural-language request

vplot_ask() is a thin, model-agnostic convenience: it assembles the grounding payload (the JSON spec_schema() plus spec_fields() for data and the user's prompt), hands it to a responder function you supply — any function that takes the payload and returns a JSON spec string (wrapping the Claude API, an MCP client, a local model, or a canned fixture) — and validates the result with vplot_from_spec().

Usage

vplot_ask(prompt, data, responder, env = globalenv())

Arguments

prompt

A natural-language description of the desired plot.

data

The data frame to plot.

responder

A function function(payload) -> json_string. payload is a list with prompt, fields (from spec_fields()), and schema.

env

Environment channel expressions are re-quoted in.

Value

A PlotSpec.

Details

vellumplot deliberately ships no built-in model client: the value is the schema, the grounding, and the validated repair loop, not a hard dependency on one provider. For a ready-to-run agent integration, point an MCP client at the bundled server (system.file("mcp/server.R", package = "vellumplot")).

See also

spec_fields(), vplot_from_spec(), spec_schema()

Examples

# a canned responder standing in for a model call:
responder <- function(payload) {
  spec_to_json(vplot(mtcars) |> mark_point(x = wt, y = mpg))
}
p <- vplot_ask("scatter of weight vs mpg", mtcars, responder)