RLM PRO

Analyze any corpus of unstructured data using Recursive Language Models - enables LLMs to handle near-infinite context through recursive decomposition.

Works with codebases, document collections, research papers, logs, and any text-based data corpus.

Based on the RLM research from MIT OASYS lab.

Installation

# Using npx (recommended - auto-installs Python package from GitHub)
npx @superadnim/rlm-pro ./my-project -q "Explain the architecture"

# Or install globally
npm install -g @superadnim/rlm-pro

Prerequisites

• Node.js 18+
• uv (Python package manager) - will be prompted to install if missing
• OpenAI API key (set as environment variable)

# Install uv if not already installed
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install the Python package (auto-installed on first run, or install manually)
uv pip install git+https://github.com/CG-Labs/RLM-PRO.git

# Set your API key
export OPENAI_API_KEY="your-key"

Usage

Command Line

# Basic usage
npx @superadnim/rlm-pro ./my-project -q "Explain the architecture"

# Get JSON output (for programmatic use)
npx @superadnim/rlm-pro ./my-project -q "List all API endpoints" --json

# Use a specific model
npx @superadnim/rlm-pro ./my-project -q "Find potential bugs" -m gpt-5.2

# Use Anthropic backend
npx @superadnim/rlm-pro ./my-project -q "Review this code" -b anthropic

# Verbose output for debugging
npx @superadnim/rlm-pro ./my-project -q "How does authentication work?" -v

# Only build context (no LLM call)
npx @superadnim/rlm-pro ./my-project -q "" --context-only

Programmatic Usage (Node.js)

const { analyzeCodebase } = require('@superadnim/rlm-pro');

async function main() {
  const result = await analyzeCodebase('./my-project', {
    query: 'Summarize the codebase structure',
    backend: 'openai',
    model: 'gpt-5.2',
  });

  console.log(result.response);
  console.log('Execution time:', result.execution_time, 'seconds');
}

main();

Options

| Option | Description | Default | |--------|-------------|---------| | -q, --query <query> | Question or task to perform (required) | - | | -b, --backend <backend> | LLM backend (openai, anthropic, etc.) | openai | | -m, --model <model> | Model name | gpt-5.2 | | -e, --env <env> | Execution environment (local, docker) | local | | --max-depth <n> | Maximum recursion depth | 1 | | --max-iterations <n> | Maximum iterations | 30 | | --max-file-size <bytes> | Max size per file | 100000 | | --max-total-size <bytes> | Max total context size | 500000 | | --no-tree | Exclude directory tree from context | - | | --json | Output as JSON | - | | -v, --verbose | Enable verbose output | - | | --context-only | Only output built context | - |

Environment Variables

| Variable | Description | Required | |----------|-------------|----------| | OPENAI_API_KEY | OpenAI API key | Yes (for OpenAI backend) | | ANTHROPIC_API_KEY | Anthropic API key | For Anthropic backend |

How It Works

RLM (Recursive Language Models) enables LLMs to handle near-infinite context by:

1. Context Building: Intelligently reads and formats your codebase
2. Recursive Decomposition: Breaks complex queries into manageable sub-tasks
3. Code Execution: Runs Python code in a sandboxed environment to explore and analyze
4. Iterative Refinement: Continues until a complete answer is found

This allows answering complex questions about large codebases that would exceed normal context limits.

Examples

Architecture Analysis

npx @superadnim/rlm-pro ./backend -q "Describe the system architecture and key design patterns"

Bug Finding

npx @superadnim/rlm-pro ./src -q "Find potential security vulnerabilities" --json

Documentation Generation

npx @superadnim/rlm-pro ./api -q "Generate API documentation for all endpoints"

Code Review

npx @superadnim/rlm-pro ./feature-branch -q "Review this code for best practices"

License

MIT

Links

• RLM Research Paper
• GitHub Repository