Create a Recursive Language Model (RLM) Module

Factory function to create an RLMModule that enables LLMs to programmatically explore large contexts through a REPL interface. Use run() to execute it. run_async() supports factory-backed modules; async streaming and module $stream() reject RLM. run_stream() preserves the synchronous forward() fallback unless a matching token-stream request is active; that request is rejected first.

Usage

rlm_module(
  signature,
  runner = NULL,
  max_iterations = 20L,
  max_llm_calls = 50L,
  max_output_chars = 100000L,
  sub_lm = NULL,
  verbose = FALSE,
  tools = list(),
  max_iters = NULL,
  ...,
  interpreter_factory = NULL
)

Arguments

signature

A Signature object or string notation defining inputs/outputs

runner

Optional caller-owned code runner implementing execute() and policy(). It is retained, never automatically closed, and must not be shared concurrently when persistent.

max_iterations

Maximum REPL iterations before fallback (default 20)

max_llm_calls

Maximum recursive LLM calls allowed (default 50)

max_output_chars

Maximum characters per execution output (default 100000)

sub_lm

Optional ellmer Chat for recursive queries. NULL = disabled.

verbose

Logical. Print execution progress (default FALSE)

tools

Named list of user-defined host functions. Guest code emits an authenticated request, dsprrr invokes the original function in the host, and the guest is replayed with the response. Closures are never deparsed or serialized into generated code.

max_iters

DSPy 3.3-compatible alias for max_iterations. Supply only one of these arguments.

...

Additional arguments passed to the module

interpreter_factory

Optional zero-argument function returning a fresh runner with execute(), policy(), optional start(), and idempotent terminal shutdown() or close(). Supply exactly one of runner and interpreter_factory.

Value

An RLMModule object

Examples

if (FALSE) { # \dontrun{
runner <- r_code_runner(timeout = 30)
rlm <- rlm_module("question -> answer", runner = runner)
result <- run(rlm, question = "What is the 10th Fibonacci number?", .llm = llm)
} # }