article research

Recursive language models: abstraction and implementation

Recursive Language Models treat prompts as external variables the model can inspect, decompose, and recursively process through a REPL, enabling context offload and programmatic subcalls. The research paper establishes the paradigm; Prime Agent implements it with a persistent IPython kernel and native subagent sessions, but depth and cost remain bounded.

v v0.7.1reviewed 2026-08-09evidence officialsources S033, S032, S008, S009cutoff 2026-08-09
Home

First-party announcement, documentation, release, or installer.

Paper abstraction

Recursive Language Models treat prompts as external variables the model can inspect, decompose, and recursively process through a REPL, enabling context offload and programmatic subcalls. The research paper establishes the paradigm; Prime Agent implements it with a persistent IPython kernel and native subagent sessions, but depth and cost remain bounded.

  • S033 supports the route's current claim set.
  • S032 supports the route's current claim set.
  • S008 supports the route's current claim set.
  • S009 supports the route's current claim set.

Reference implementation

Recursive Language Models treat prompts as external variables the model can inspect, decompose, and recursively process through a REPL, enabling context offload and programmatic subcalls. The research paper establishes the paradigm; Prime Agent implements it with a persistent IPython kernel and native subagent sessions, but depth and cost remain bounded.

  • S033 supports the route's current claim set.
  • S032 supports the route's current claim set.
  • S008 supports the route's current claim set.
  • S009 supports the route's current claim set.

Prime Agent mapping

Recursive Language Models treat prompts as external variables the model can inspect, decompose, and recursively process through a REPL, enabling context offload and programmatic subcalls. The research paper establishes the paradigm; Prime Agent implements it with a persistent IPython kernel and native subagent sessions, but depth and cost remain bounded.

  • S033 supports the route's current claim set.
  • S032 supports the route's current claim set.
  • S008 supports the route's current claim set.
  • S009 supports the route's current claim set.

Depth and cost

Recursion depth is configurable but bounded. Use the child registry and message limits to keep costs predictable.

What this does not mean

This feature or concept does not imply a sandbox, a guarantee of success, or a claim that the harness is superior in every scenario. It only means the documented behavior exists in v0.7.1.