An agent that remembers what you told it last week
Hermes is an agent platform rather than a chat window. It keeps memory between sessions, schedules work for later, hands jobs to sub-agents and reaches real tools to browse, read files and run code. Hosted by us, all of that sits on a machine you control.
- Memory across sessions
- Scheduled and delegated work
- Your model, your choice
- Private to your server
- 1MondayYou asked what the supplier agreed about lead times.
- 2TuesdayIt read the contract and the meeting notes, and kept what it found on your disk.
- 3ThursdayYou asked again. It answered from memory, and named both sources.
The memory lives on your disk. A new session does not start from nothing, and nothing is stored with a third party.
The parts that make it an agent rather than a chatbot
Memory that persists
Sessions build on each other instead of starting cold, and the memory file lives on your disk.
Scheduled work
Tasks can be set for later or on a recurrence, and the agent carries them out unprompted.
Sub-agents
Long jobs are delegated and run in parallel, which is why machine size matters here.
A real tool layer
Browsing, file handling and code execution, rather than answers alone.
Swap the model underneath
The agent and the model are separate; change one without rebuilding the other.
Reachable only by you
The interface is served over an SSH tunnel, and it can sit behind a private network so only your devices reach it.
The machines it runs on
Every plan is a virtual machine of your own with root access, billed monthly in rupees. The application is installed before we hand it over.
How much machine you need depends on how many sub-agents run at once. One person with one agent is light work; a team with several agents working in parallel is not.
Basic 12 is the size to pick for a team: enough memory and cores for several sub-agents at once, with room for the memory store to grow.
Prices are for the server. The agent is open source; the model behind it is your choice and your cost, whether that is an API key with a provider or a local model running on the same machine. Basic 4 has a one-time setup fee on a monthly contract, shown beside its price; there is none on yearly billing.
Run it against an API, or keep everything local
The agent does not come with a model, and that is a feature. Point it at a commercial API when quality matters most, or at a model running on the same machine when nothing may leave your infrastructure.
- A provider API key: the strongest models, billed per token by them
- A local model on the same server: no per-token bill, no data leaving the machine
- Both, with different agents pointed at different models
- Change your mind later without rebuilding the agent or losing its memory
- We will tell you honestly what a local model can and cannot do on a machine without a GPU
Worth knowing
No GPU on these machines
Our VPS plans are CPU only. Large local models will be slow; a commercial API or a small local model is the realistic choice here.
The agent is installed, the model is not
We build the machine with the agent ready. Pointing it at a model and connecting your tools is a step you take over SSH.
It is unmanaged
We support the server and the install. The agent's configuration, updates and behaviour are yours, like any root-access machine.
Size it for sub-agents
One agent answering questions is light. Several sub-agents working in parallel is the load that decides your plan.
We also build agents for businesses, end to end
Self-hosting suits people who want the machine and the control. If what you actually want is an agent connected to your order system, answering customers and escalating properly, that is a project our AI team takes on — built, measured and supported.
See AI agents and automationOther things we host and build
Hermes AI Hosting, answered
What exactly is Hermes?
Two things with one name. Hermes is a family of open-weight language models, and the Hermes agent is the platform built around them: memory, scheduling, sub-agents and tools. What we install is the agent.
What exactly is Hermes?
Two things with one name. Hermes is a family of open-weight language models, and the Hermes agent is the platform built around them: memory, scheduling, sub-agents and tools. What we install is the agent.
Give the agent a machine and a memory
Choose a size and a region, and we will build the server with the agent installed and ready to be pointed at your model.