# rlm

> Process a query using RLM (Recursive Language Model) for huge contexts that exceed normal limits. Use when working with very large documents, codebases, or datasets.

- Author: RossRemoteSense
- Repository: remotesense/recursive-llm
- Version: 20260127210426
- Stars: 0
- Forks: 0
- Last Updated: 2026-02-06
- Source: https://github.com/remotesense/recursive-llm
- Web: https://mule.run/skillshub/@@remotesense/recursive-llm~rlm:20260127210426

---

---
name: rlm
description: >
  Process a query using RLM (Recursive Language Model) for huge contexts
  that exceed normal limits. Use when working with very large documents,
  codebases, or datasets.
disable-model-invocation: true
argument-hint: [context-file] [query]
---

# RLM Query Command

You are processing a query using Recursive Language Model techniques.

## What This Does

RLM allows you to answer questions about contexts that are too large to fit in
a normal context window. It works by:

1. Storing the context in a file (not in the model's context)
2. Using code to explore, filter, and chunk the context
3. Delegating semantic analysis to the rlm-analyzer subagent
4. Aggregating results to form the final answer

## How to Use

Invoke with: `/rlm <context-file> <query>`

Examples:
- `/rlm /path/to/huge_log.txt "How many errors occurred on January 15th?"`
- `/rlm ./data.json "Summarize all entries with status=failed"`

## Process

### Step 1: Validate Input

Parse the arguments:
- First argument: path to context file
- Remaining arguments: the query

If no file path provided, ask the user for the context file location.

### Step 2: Prepare Context

Copy/link the context file to a working location:
```bash
cp "$ARGUMENTS" /tmp/rlm_context.txt 2>/dev/null || echo "Provide context file path"
```

### Step 3: Invoke RLM Processor

Delegate to the rlm-processor subagent:

```
Use the rlm-processor subagent to answer this query:

Context file: /tmp/rlm_context.txt
Query: <extracted query>

Process this using RLM techniques - peek at the structure first, then use
appropriate strategies (grep/filter, chunk+map, or aggregation) based on
the query type.
```

### Step 4: Return Result

The rlm-processor will return the final answer. Present it to the user along with:
- How many chunks were processed (if applicable)
- Any relevant metadata about the processing

## Notes

- RLM is for LARGE contexts (>100k tokens) that need semantic understanding
- For simple grep-able tasks, direct bash commands may be faster
- The rlm-patterns skill has additional processing strategies
- Use rlm-analyzer subagent for semantic analysis of chunks