How Piskes Agent differs from cloud-run AI agents
The biggest difference between Piskes Agent and a cloud-run AI agent isn’t features — it’s where the agent runs. Piskes Agent runs on your own Windows machine, using a model key you configure yourself; a cloud-run agent runs on the provider’s servers, staying online even when your machine is off. Each direction has real trade-offs — neither wins outright.
Piskes Agent: local-first, your own key
- The Hermes Agent engine runs on your machine; files, bot profiles and memory live in
~/.hermeson your own disk, not on a Piskes server. - You choose your model provider and pay them directly at their listed price (BYOK) — Piskes doesn’t sit in between unless you sign in to use hosted quota.
- Trade-off: scheduled bots (weekly reports, Monday briefings…) only run while your machine is on and the app is running in the background. Turn your laptop off for a trip and that day’s run simply doesn’t happen — you’ll see it when you turn it back on.
Cloud-run agents: always online, data through a third party
- The agent runs on the provider’s infrastructure 24/7, independent of whether any personal machine is powered on.
- In exchange, the data the bot handles (email, files, company numbers) has to pass through that provider’s servers for the agent to work — you’re trusting their security and retention policy rather than controlling it directly.
- Usually billed per agent per month, or by usage set by the provider — harder to separate a “platform fee” from a “model fee” the way BYOK does.
So which one fits
| Priority | Better fit |
|---|---|
| Sensitive company data, want to control where it’s stored | Local-first (Piskes) |
| Need a process running truly 24/7, independent of any personal machine | Cloud-run agent |
| Want to choose and switch models, paying the provider’s listed price | Local-first, your own key |
| Multiple people sharing usage, want Piskes to handle model quota instead of managing keys yourself | Piskes hosted plan (Starter/Pro/Team) |
Piskes Agent defaults to local-first because most office work (drafting documents, compiling spreadsheets, screening resumes) doesn’t need to run 24/7 — running when you ask, or on a business-hours schedule, is enough, and in exchange your data stays on your own machine. Even the hosted plan only changes who pays for the model: the Hermes Agent engine still runs on your machine, it doesn’t move to a Piskes server. If the job genuinely requires a process that keeps running independent of any personal machine, that’s a problem a cloud-run agent solves better.
Further reading: Configuring a model provider · How pricing and fees work · Pricing.