A Linux VPS hosted in Montreal, Canada, with Hermes Agent, the open source AI agent from Nous Research, installed and started on delivery. It runs around the clock on your machine, keeps its memory from one session to the next, and you talk to it from Telegram, Discord, Slack or Signal.
Your Hermes Agent VPS is delivered from our Montreal datacenter: four tiers, Intel Xeon Gold 6132, DDR4 ECC and NVMe, from €7.90 a month.
The datacenter you pick does not change how fast the agent is: that depends on the model provider you plug into. What it does change is where its memory is kept. Here, everything your agent has remembered is stored in Canada, with no copy abroad. Enough to give a straight answer to the client handing you their files.
Your Hermes Agent VPS runs on Intel Xeon Gold 6132 processors, error-correcting DDR4 ECC memory and 100% NVMe storage, in our Montreal data center. DDoS protection included and support 7 days a week on all four tiers, from €7.90 to €32.90 a month.
The agent’s memory, its skills and its configuration are written in Montreal and do not leave. No replication to another country, no clause to negotiate: you answer in one sentence the client or the legal team that requires everything to stay in Canada.
And your users gain too: Montreal puts your instance a few tens of milliseconds from Quebec, Ontario and the north-eastern United States. Order in a few minutes, Hermes Agent is installed and started on delivery.
Hermes Agent is the open-source AI agent from Nous Research, under an MIT licence. It runs as a daemon on your machine and is driven from more than sixteen messaging apps: Telegram, Discord, Slack, WhatsApp, Signal, email. Elsewhere, allow a quarter of an hour to install it. At NorthHost it is already done by the time you get your credentials.
Most agents forget everything between two conversations. Hermes Agent keeps its memory and writes its own skills: explain a procedure once, and it recalls it on its own. All of it as files, on your disk, that you read, correct and back up. Change model provider and it loses nothing.
The reasoning happens at the model provider you plug in. Your server orchestrates, remembers and executes. An agent calling Claude therefore runs from €7.90 a month, where a machine able to host the model would cost ten times that. Local inference, on the other hand, is not for these processors.
Hermes 4, Claude, GPT or any compatible provider: the engine is separate from the model, you switch without touching the code and without losing the memory. The software is free. You pay for the VPS from €7.90 a month, and for the model by usage. That second line is the one to watch.
Edge, 6 vCore, 8 GB and 110 GB of NVMe at €12.90 a month, is the tier we recommend: the agent’s memory grows month on month, the skills pile up and the integrations multiply as soon as it watches several messaging apps. Orion and Legacy for the busiest agents. Change tier pro rata, without reinstalling a thing.
A Linux VPS with Hermes Agent, the open-source AI agent from Nous Research, installed and started on delivery. It runs continuously on your machine, keeps its memory from one session to the next, and you talk to it from Telegram, Discord, Slack or Signal.
An online assistant starts from nothing at every conversation, and whatever it understood about your trade stays with its vendor. Hermes Agent does the opposite: its memory accumulates from one session to the next and is written to the disk of your server. After a few weeks, that disk holds a knowledge of your organisation that exists nowhere else.
It goes further: from what it has learned, it writes its own reusable skills. They are files, on your machine, that you can read, correct and back up. That is the point of self-hosting here: this kind of capital is not something you rent.
A memory that builds up: From one session to the next, Hermes Agent remembers the context of your work instead of asking for it all again.
It writes its own skills: From what it has learned, the agent writes reusable skills that it later recalls on its own.
The engine, not the model: Hermes 4, Claude, GPT or any other compatible provider: you change model without touching the code.
MIT licence: The software is free and permissive. You pay only for the server and the model provider you plug in.
An agent that runs continuously, remembers, and is driven from the messaging apps you already use.
Persistent memory: The accumulated context survives closing the conversation and restarting the server: it is on the disk.
Self-written skills: The agent turns what it has learned into reusable skills, which it later recalls without being asked.
Over 16 messaging apps: Telegram, Discord, Slack, WhatsApp, Signal, email and the command line: you talk to the agent where you already are.
Scheduled tasks: Cron-style scheduling: the agent wakes up at the set time and does what was planned, with no nudge.
Persistent daemon: Hermes Agent does not run for the length of a command: it runs continuously on your infrastructure.
The model of your choice: The engine is separate from the model. Hermes 4, Claude, GPT or any compatible provider, without editing the code.
Open source, MIT licence: Released in February 2026 by Nous Research, currently at version 0.10.0. The code is public and the licence is permissive.
Root access: Hermes Agent is added to your VPS, it does not replace it. Full SSH to install whatever you want alongside.
Our Hermes Agent VPS are hosted in Montreal, Canada. The agent's memory, its skills and its configuration stay on Canadian soil, with no replication to another country.