hermes-lcm: Lossless Context Management

hermes-lcm is a plugin for Hermes Agent that introduces a Lossless Context Management (LCM) system. It addresses the issue of context loss in active conversations by persisting conversation history and then compacting older turns into a summary Directed Acyclic Graph (DAG). This DAG allows the agent to efficiently retrieve relevant information from past interactions without losing crucial details. The plugin utilizes a SQLite message store, a Summary DAG, and agent tools for granular context retrieval.

This project differentiates itself through its explicit focus on context retrieval quality and autonomy. Unlike Hermes' built-in compression, hermes-lcm provides a plugin-local store and a drill-down mechanism using LCM tools. It offers source-aware retrieval, session controls (including ignore/read-only), and large payload handling, offering a more controlled and adaptable context management solution.

• Message Persistence: Preserves raw messages before compaction for full recovery.
• Summary DAG: Creates a depth-aware summary of older context for efficient retrieval.
• Agent Tools: Provides tools like lcm_grep, lcm_describe, and lcm_expand for precise context access.
• Source Lineage: Tracks the source of each message and summary for targeted retrieval.
• Session Controls: Enables filtering sessions based on patterns or content.
• Large Payload Handling: Supports externalization of oversized tool results and media.
• Sensitive Pattern Controls: Offers redaction of sensitive information before storage.

The project is actively developed and maintained, with recent commits and ongoing issue resolution. Comprehensive documentation is provided, and the plugin has a growing community. The plugin has passed CI checks and is regularly updated, indicating a stable and reliable codebase. Regular updates suggest ongoing improvements and support for new features.

hermes-lcm benefits developers building applications with Hermes Agent by providing a robust and flexible context management system. It's ideal for use cases requiring precise control over context retention, efficient information retrieval, and handling of large payloads. By offering a lossless recovery mechanism and granular tools, it enhances the autonomy and reliability of Hermes Agent applications, providing more robust and predictable conversational experiences.