Tokenjuice Trims Hermes Agent's Terminal Noise Before the Model Reads It

A new plugin for Hermes Agent cuts long terminal output down before the AI model reads it. Tokenjuice joined the official Hermes plugin catalog on the 2nd of November 2026. It is maintained by Vincent Koc and is free under the MIT licence.

Each time Hermes runs a command, it passes the whole printout to the model, and providers bill by the token. Tokenjuice shortens the long ones on your own machine first, by fixed rules, and the catalog page says the plugin sends nothing over the network.

On the 4th of November, the unofficial tips account HermesAgentTips posted a test on Hermes 0.22.6. A 700 line folder listing went from 44,082 characters to 774. A short "git status" check went from 670 to 223.

Only commands that worked get trimmed. Failed commands, background jobs, exact file reads and very large outputs reach the model in full, according to the plugin page. It needs Hermes 0.22.5 or newer and Node.js 22 or newer, and it is a community plugin, reviewed for the catalog but not written by Nous Research.

Read the step by step guide