Aleph

Implements Recursive Language Models (RLMs) for unbounded context processing by letting LLMs programmatically interact with large documents through a sandboxed Python REPL environment. Rather than solving context limits at the architecture level, Aleph enables models to partition context and make recursive calls to themselves, using tools like search, peek, lines, chunk, sub_query, and sub_aleph to iteratively explore and reason over massive inputs. Supports both standalone usage and MCP server mode with full trajectory logging and provenance tracking, making it ideal for document analysis, multi-document QA, and complex reasoning tasks that exceed traditional context windows.