# witcheer's Wingtips #2: the 🗜️ icon, what context compression keeps and what it drops

Written by [witcheer](https://x.com/witcheer) at Nous Research. Republished with credit.

Canonical: https://hermeswatcher.com/wingtips/2  
Published: 2026-06-19  
[Original](https://notwitcheer.github.io/hermes-recipes/wingtips/02-the-icon-what-context-compression-ke/) · [On X](https://x.com/witcheer/status/2068027535955468533) · [Full collection](https://notwitcheer.github.io/hermes-recipes/) · [Official docs](https://hermes-agent.nousresearch.com/docs/developer-guide/context-compression-and-caching)

[Source Markdown](https://raw.githubusercontent.com/notwitcheer/hermes-recipes/8653fb699a137a32e81936fa33cabacd3f4f2095/docs/wingtips/02-the-icon-what-context-compression-ke.md): use this file as the base for relative links in the verbatim text below.

Sources: [Hermes docs: Context Compression And Caching](https://hermes-agent.nousresearch.com/docs/developer-guide/context-compression-and-caching)


Hermes Wingtips #2

the 🗜️ icon, what context compression keeps and what it drops

this little emojy is the compression count: how many times Hermes auto-summarised the session to stay under the context limit, which kicks in around 50% full. 

when compression occurs, it keeps your first 3 turns and your last 20, and summarises everything in between. 
a detail from the middle of a long session can then drop out, and the agent repeats work it already did, even though the opening goal and the recent turns are intact.

~~~
three levers when it bites, all in config.yaml, hot-reloading on a running gateway: 

- protect_last_n keeps more recent turns uncompressed 

- auxiliary.compression.model points the summariser at a cheap fast model so it does not burn main-model tokens

- model.context_length raises the ceiling so it fires later.

https://hermes-agent.nousresearch.com/docs/developer-guide/context-compression-and-caching