Ruby Rails Documentation Generator

A DSPy program that generates documentation for Rails applications using Recursive Language Models (RLMs). Created by Drew Breunig

What are RLMs?

Recursive Language Models are an inference strategy where language models can decompose and recursively interact with input context of unbounded length through REPL environments. Instead of processing huge contexts in a single call, an RLM allows the model to programmatically partition, grep, and recursively sub-query the context—avoiding context rot while handling arbitrarily large inputs.

Key benefits:

Near-infinite effective context length
Reduced context rot degradation
Cost-efficient processing of large codebases

For more details, see the RLM paper.

Installation

uv add modaic dspy

Usage

Load with AutoProgram

from modaic import AutoProgram

# Load the program from Modaic Hub
doc_writer = AutoProgram.from_precompiled("farouk1/ruby-rails-doc-generator")

# Generate documentation for a Rails project
result = doc_writer(source_root="./my-rails-app")
print(result.documentation)

Override Config

from modaic import AutoProgram

doc_writer = AutoProgram.from_precompiled(
    "farouk1/ruby-rails-doc-generator",
    config={"max_iterations": 15, "lm": "openai/gpt-5"}
)

Load a Specific Revision

from modaic import AutoProgram

# By branch
doc_writer = AutoProgram.from_precompiled("farouk1/ruby-rails-doc-generator", rev="dev")

# By tag
doc_writer = AutoProgram.from_precompiled("farouk1/ruby-rails-doc-generator", rev="v1.0")

# By commit hash
doc_writer = AutoProgram.from_precompiled("farouk1/ruby-rails-doc-generator", rev="918ad95")

Configuration

Parameter	Default	Description
`max_iterations`	10	Maximum RLM iterations
`lm`	`openai/gpt-5-mini`	Primary language model
`sub_lm`	`openai/gpt-5-mini`	Model for recursive sub-calls

License

MIT

2.0 KiB Raw Permalink Blame History