Does headroom work with Hermes?

[Zero1man]

I am using GPT 5.5 as the base model in Hermes. Does this work with Hermes?

[hosanxiv]

Yes. Hermes can use Headroom when its model traffic is routed through Headroom's OpenAI-compatible proxy. Using GPT-5.5 as the base model does not by itself prevent this; the important part is that the configured provider uses an OpenAI-compatible endpoint and request format.

The current setup is:

1. Start headroom proxy with your existing OpenAI-compatible upstream.
2. Point Hermes's custom provider/base URL at http://127.0.0.1:8787/v1.
3. Keep the API key and model name expected by the upstream provider.

For reversible compression, also install the repository's plugins/hermes/headroom_retrieve plugin and enable it in ~/.hermes/config.yaml. The Hermes plugin documentation recommends excluding file reads and retrieval itself:

HEADROOM_EXCLUDE_TOOLS=read_file,headroom_retrieve

That avoids compressing file reads and prevents a retrieve/recompress loop. The general manual proxy configuration is also documented here.

One caveat: the plugin README mentions headroom wrap hermes, but the current CLI does not appear to expose Hermes as a dedicated wrap subcommand. Starting the proxy explicitly and setting Hermes's base URL is therefore the clearest path with the current release.