Hermes Agent
One assistant, 200+ brains to choose from — and the memory is always yours.
- Free local per message or API with your key
- Switch models without losing memory
- Control panel + WhatsApp support
Managed plan · no lock-in
Get my Hermes Hermes · AI models
200+ models in Hermes: which to use in each situation
Hermes is model-agnostic: the assistant is one (with your memory) — the brain is pluggable.
TL;DR
The Hermes Agent supports 200+ AI models. Rule of thumb: your key as the default (GPT, Claude, Gemini or Grok, paid directly to the provider) and API as the exception (GPT/Claude with your key, for heavy tasks). Switching models does not erase memory — it belongs to Hermes, not to the model. On Rollin Host, everything managed from the panel.
The 30-second decision
Daily use (memory, audio summaries, reminders, drafts, organization): local model. Zero cost per message, data that never leaves the server, great speed for conversation. It is the right default for 90% of personal use.
Heavy tasks (long document analysis, code, complex reasoning): plug in a model via API with your key — GPT, Claude, Gemini — only when needed. You pay the provider for usage, and Hermes remains the same assistant, with the same memory.
Local vs API, in practice
| Local model (separate LLM server) | API model (GPT/Claude) | |
|---|---|---|
| Cost | Zero per message (just the VPS) | Pay-per-use to the provider |
| Privacy | Nothing leaves the server | Data passes through the provider |
| Power | Great for conversation and organization | Superior on complex tasks |
| When to use | Day-to-day default | Exception, on demand |
| Hermes memory | Kept | Kept (it belongs to the agent, not the model) |
Read next
Frequently asked questions
Which AI models does Hermes support?
Over 200 — in two families: LOCAL models (Llama, Mistral, Qwen and the like, running on a separately contracted LLM server) and API models (GPT, Claude, Gemini — you connect them with your key and pay the provider for usage).
Which model should I start with?
A budget model from the provider you already use (GPT mini, Gemini Flash or Claude Haiku), with your key: for memory, summaries, reminders and drafts it does the job for cents. Move up to a bigger model when the task calls for more power — long analyses, code, heavy reasoning.
Does switching models erase memory?
No. Memory and learning belong to HERMES, not to the model — the model is the pluggable engine. You swap brains while keeping everything the assistant knows about you.
Is a local model worth it?
Only if you have a separate LLM server; it is not included in Hermes. For personal use, your own key from a provider is cheaper and stronger: budget models cost cents a day.
Pronto pra hospedar seu projeto de IA?
Comece em 5 minutos. Migração gratuita (no plano semestral), suporte 24/7 em português e garantia de reembolso de 7 dias (30 dias em hospedagem de sites e WordPress).
