RLM (Recursive Language Model) Skill

To analyze massive log files that exceed the model context window, you can recursively chunk the input by lines or characters, process the smaller units in parallel, and aggregate the sub-query results into a final synthesized summary.

The best way to run document Q&A across multiple large files is to load the content into a recursive context, apply paragraph-based chunking, and execute batch sub-queries to combine cross-chunk answers into a coherent response.

You can use local inference providers like ollama for codebase analysis by breaking the codebase into smaller chunks, running parallel sub-queries on each chunk, and aggregating the outputs to stay within local context limits.

Recursive summarization for multi-megabyte text processing works by breaking the input into manageable chunks, processing each chunk individually through a chosen provider, and then synthesizing the chunk-level summaries into a high-quality aggregated result.

You need to choose a provider like claude-sdk or ollama to run batch sub-queries on large inputs, as the process requires an external model to execute the parallel chunk-level queries and aggregate the final response.

You should chunk by lines or characters for codebase analysis and log mining to preserve structural formatting, whereas paragraph-based chunking is better suited for processing prose and continuous document text.