Thank you, @Sondra Patton . What is the best model setup you are currently using to maximise Hermes without spending much?

Update: how I got it running on my Claude subscription instead of the API Big credit to Dmitry's post on using auth tokens, which pointed me the right way. It behaves differently depending on your account though, so here's what I found. Why it works for some people and not others. `CLAUDE_CODE_OAUTH_TOKEN` bills to your 'extra usage credits', not the usage included in your plan. If those are enabled and funded, it just works. If not, you get `400 You're out of extra usage`. That was my wall for weeks. Not a config mistake, just which pot the request comes out of. What worked instead. Have Hermes run the Claude Code command in the terminal: 'claude -p "your prompt" --model sonnet' Because that runs as Claude Code, it comes out of your plan's included usage. I checked the real billing. Zero API spend. Two things that will bite you. 1. You must have no `ANTHROPIC_API_KEY` in your environment. Hermes loads its `.env` into its own process and anything it launches inherits that. So if a key is sitting there, the CLI quietly bills the metered API instead of your subscription. Nothing errors, and your Hermes cost report won't show it. 2. `claude -p` returns exit code 0 even when it fails. It says `subtype: "success"` too. The truth is in `is_error` and `terminal_reason`. Check the exit code and you will eventually hand yourself the error message as your answer. One bonus. Run it with `--tools ""` so Claude only thinks and Hermes keeps the tools. Roughly seven times cheaper on usage: the full agent setup carries about 24k tokens a call against 3.5k. Every answer I read now comes from Sonnet, on the subscription I was already paying for.