stephenschoettler/hermes-lcm
hermes-lcm is a lossless context management plugin for Hermes Agent, providing a DAG-based context engine that ensures no messages are lost through persistent storage and intelligent summarization.
Awesome Hermes Plugins › Context Engines
Sahil-SS9/hermes-memlock
MemLock addresses a specific failure of long Hermes sessions: when context is compacted, earlier turns are wrapped in a marker that tells the model everything below it is background reference rather than active instruction. A standing instruction about output format, tool preference or writing style that ends up inside that summary region is still physically present but has been demoted, so it quietly stops being followed. MemLock lets the user pin such instructions as anchors, detects each compaction, audits which anchors survived in the active non-summary region, and re-injects the casualties as a reminder block before they drift out of the model's working memory. Pins are created through a `guard_pin` tool with a text and a priority, can be updated in place under the same identifier while the old version is kept, and current state is inspected with a `/guard` command. A stdlib-only setup script replaces hand-editing of configuration: it probes the Hermes installation, reports every finding, and writes only the `memlock:` section of `~/.hermes/config.yaml` after taking a timestamped backup, leaving everything outside that section untouched. Detection covers preference-aware memory backends — a Severian plugin directory or `SEVERIAN_DSN` variable, a Mnemosyne plugin directory or enabled-config entry — and falls back to a provider-free mode where core pinning and auditing still work and only the reverse audit pass stays off. The script supports a dry run, an explicit provider override and a pruning option that deletes session store files older than a given number of days by modification time, with a minimum of seven days, never touching durable global pins or non-JSON files. Beyond Hermes, where it runs as a native plugin with full hook support, the design carries to Claude Code through settings-file hooks that snapshot before compaction and audit after, and to generic MCP clients through manual tool calls exposing pin, unpin, update, status and audit operations.
https://github.com/Sahil-SS9/hermes-memlock
Also in Context Engines
claudioemmanuel/squeez
squeez is an end-to-end token optimizer that provides hook-based token compression for Hermes Agent and several other AI CLI hosts, increasing context window efficiency.
EfficientContext/ContextPilot
ContextPilot is a context engine that accelerates long-context LLM inference by optimizing context reuse, offering significant speedups and token savings for RAG, memory chat, and agentic AI worklo...
byenzyme/enzyme-skill
A Hermes plugin that compiles a Markdown vault into a local concept graph and injects trending entities and semantic search results into the session through three lifecycle hooks and five tools.
beniedev/moonbite
Experimental Hermes Agent plugin adding cross-session memory, short-lived working state, and bounded, verified autonomous action on top of the host agent.
eas4ai/hermes-context-manager
Silent-first context optimization plugin for the Hermes Agent gateway that compresses tool outputs, deduplicates repeated work and summarizes completed phases without the main model noticing.
kenyonxu/hermes-persona
hermes-persona is a configurable Hermes Agent plugin that dynamically injects persona context into LLM calls using various hooks and rules.
PlayForm/Aphrodite-Hermes
Compresses Hermes tool output before it reaches the model into compact retrievable previews, via a thin Python loader calling a Rust classifier and store, cutting per-call token cost.