Published: 2026-09-25
Summary

Hermes desktop's new features, verified: Bot Mode, HUD, voice, "Hey Hermes" and cron continuity

Chapters / key moments (click to jump — plays here on the page)

Alex Finn walks through the Hermes desktop features he uses daily and gives one strong opinion: make one Bot per model, not one Bot per job title. We checked each feature against the Hermes desktop docs. One of his claims needs a correction: cron jobs don't remember past runs until you turn that on.

Source video

"The newest Hermes agent update is unbelievable" by Alex Finn — Watch on YouTube →

Bot Mode: one Bot per model

Per the docs, Bot Mode is built into the desktop app and on by default (a Bots tab next to Sessions). Each Bot is simply a Hermes profile, with its own model, memory, skills and chat history under ~/.hermes/profiles/<name>/. That means the CLI sees the same Bots:

hermes -p <bot> chat

His advice: Hermes' strength is running any model from any provider, so build Bots around models rather than roles. For example, a GPT-6 Bot for building, a Claude Bot for front-end work, a cheap Bot for bulk coding, and a local-model Bot for simple free tasks. You can ask Hermes itself to create them ("set me up a new bot powered by Claude").

Local models from Settings

He shows Settings → Providers → run models locally, which he says checks your hardware, recommends a model, and downloads, loads and tests it before attaching it to a Bot. The docs confirm a Settings → Local models panel (llama-server) in the desktop app. The hardware-recommendation flow is as shown in the video. Small local models (under ~30B) are weaker at tool calling; see best free models for Hermes.

HUD mode

⌘/Ctrl+Shift+H detaches the chat into a chrome-free, always-on-top floating bar over whatever you're working in (documented under the desktop app's HUD mode). You can also enable Tap to summon HUD in Settings → Keyboard Shortcuts → HUD gesture. He uses it with screen context, asking about what's on screen without switching windows.

Voice and the "Hey Hermes" wake word

The composer's microphone is dictation. Hover it for Read replies aloud and the wake word toggle, and use the button to its right for a full voice conversation. The wake word is an on-device listener that opens a voice session on the CLI, TUI or desktop app. In the CLI, toggle it with:

/wake on
/wake status

He uses voice to hand work to other Bots ("Hey Hermes, tell Claude to build…"). He needed an OpenAI API key for the voice mode he demos and puts the cost at "about 5 cents a minute". That price is his estimate, not a documented one.

Subagents you can see (and re-model)

Multi-part tasks now fan out into subagents with a per-subagent window. You can change the model or skills each one uses. His tip for design work: give each subagent a different model and compare the results.

Correction: cron "memory" is opt-in

He says cron jobs now remember past runs, so daily research stops repeating itself. The docs describe this as continuity, and it is off by default: "Set continuity=true and the job injects its own most recent output into each run." Turn it on when you create the job:

hermes cron create "every day at 7am" "Research AI stocks. Skip companies covered in your previous run." --continuity

More in our Hermes tasks and cron guide.

His stack

Hermes handles multi-model work, cron research and managing his own machines ("more willing to jump between my computers"). ChatGPT Work handles deep technical work and computer use, and Grok Bot covers quick knowledge work from its mobile app. He calls Hermes' missing mobile app its biggest weakness. (Includes a sponsor segment.)