多智能体LLM编排框架

Installation

You can install mini007 from CRAN with:

 r
install.packages("mini007")

The documentation is available here

Key components

 r
library(mini007)

retrieve_open_ai_credential <- function() {
  Sys.getenv("OPENAI_API_KEY")
}

llm <- ellmer::chat(
  name        = "openai/gpt-4.1-mini",
  credentials = retrieve_open_ai_credential,
  echo        = "none"
)

`Workflow` explicit sequential pipelines

A Workflow lets you build a predefined pipeline of Stations connected by Routes. Each Station’s output becomes the next one’s input. Unlike LeadAgent, the execution path is fully explicit, you control the order and branching logic.

 r
wf <- Workflow$new(name = "article-pipeline")

wf$add_station("research", Agent$new(
  name        = "researcher",
  instruction = "Gather concise facts on the topic.",
  llm_object  = llm
))
wf$add_station("write", Agent$new(
  name        = "writer",
  instruction = "Turn the facts into an engaging paragraph.",
  llm_object  = llm
))
wf$add_station("edit", Agent$new(
  name        = "editor",
  instruction = "Polish the paragraph for grammar and clarity.",
  llm_object  = llm
))

wf$add_route("research", "write")
wf$add_route("write",    "edit")

wf$run("The history of the printing press")

Key capabilities:

- Mixed handlers: Stations accept Agent objects, WorkflowAgent objects, or plain R functions, making it easy to mix LLM calls with deterministic pre/post-processing steps. - Conditional routing: add_route(from, to, condition) gates a route on a function of the previous Station’s output, enabling branching pipelines. - Caching: with use_cache = TRUE (default), each Station’s result is stored by (name, input). Re-running the same input serves the cache instantly. Use $clear_cache() to reset. - Wrapping as an agent: $as_agent() converts any Workflow into a WorkflowAgent that can be registered with a LeadAgent or embedded as a Station inside another Workflow. - Visualisation: $visualize() renders the pipeline as an interactive directed graph via DiagrammeR.