Hermes VCC Engine

What Hermes VCC Engine does

When a Hermes conversation fills its context window, stock Hermes asks your model to write a summary of everything so far. On a local model the README reports a median of 8 minutes, up to 21, in its benchmark. This plugin replaces that step with fixed rules that copy facts from the recorded messages, finishing in about 0.05 seconds with no model call.

The summary preserves which commands ran, which failed, which errors appeared and which files were read or changed, and every line links back to the original message. It is based on the paper View-oriented Conversation Compiler for Agent Trace Analysis, arXiv 2603.29678. The README says it helps most on local or slow models and long coding or ops sessions, with a smaller gain on fast cloud APIs. Each compaction logs the projection time, and you can switch back to the stock compressor with one config command.

Key features

• Compaction summary built in about 0.05 seconds
• No LLM call during compaction
• Keeps exact commands, exit codes, errors and files
• Every summary line links back to its source message
• Switch back with hermes config set context.engine compressor
• Based on the VCC paper, arXiv 2603.29678

When to use it

• Avoiding multi-minute compaction stalls on a local model
• Keeping facts straight in long coding or ops sessions
• Checking which command failed after repeated compactions

Who it is for: Hermes Agent users on local or slow models whose long sessions stall each time the context fills.

How it fits with Hermes Agent

Built for Hermes Agent as a plugin that replaces the context engine used for compaction.

How to install Hermes VCC Engine

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

hermes plugins install danbiocchi/hermes-vcc-engine --enable
hermes config set context.engine vcc

Requirements: Hermes 0.21.7 or newer; restart Hermes after installing

FAQ