Skip to main content
Log in Sign up FREE!

Hermes on your VPS or dedicated server: self-improving AI

By ServerPoint's Team

Hermes Agent is an open-source, self-improving autonomous AI agent built by Nous Research. It’s MIT-licensed, launched earlier in 2026, and has hit over 100,000 GitHub stars by mid-April 2026. Unlike a chatbot wrapper or an IDE copilot, Hermes Agent is a persistent service that runs on your own infrastructure. A modest VPS is enough to start, and a dedicated server gives it room to grow into a serious always-on automation hub.

What makes it different: every successful complex interaction becomes a reusable “skill document” that the agent can retrieve and improve over time. The longer you use it, the better it gets at your specific work.

What Hermes Agent is

Hermes Agent is a persistent, always-on AI agent with:

  • 40+ built-in tools: web search, browser automation, code/terminal execution, image generation, TTS, vision, and more
  • Sandbox execution: local, Docker, SSH, Singularity, or Modal. You pick the isolation level.
  • Model-agnostic backend: Nous Portal, OpenRouter (200+ models), OpenAI, Anthropic/Claude, NVIDIA NIM, Xiaomi MiMo, GLM, Moonshot, MiniMax, Hugging Face, local Ollama, or any OpenAI-compatible endpoint. No lock-in.
  • Multi-interface: CLI/TUI on your server, plus gateway to Telegram, Discord, Slack, WhatsApp, Signal, and email.
  • Natural-language scheduling: cron-style jobs without cron syntax. “Every weekday at 7 AM, summarize my mentions.”
  • Built-in persistence: conversation history, skill library, and user model survive across sessions, restarts, and subagents.

The self-improving learning loop

This is the part that matters. Most agents are stateless. They plan, execute, and forget. Hermes Agent runs a five-stage loop:

  1. Execute. Plan, use tools, complete the task.
  2. Evaluate. Explicit feedback (your corrections) and implicit signals (you accepted the result) score the outcome.
  3. Extract. Non-trivial successes become structured skill documents with reasoning patterns, tool sequences, and decision points.
  4. Refine. When a better approach shows up, the agent updates existing skills.
  5. Retrieve. On future tasks, it searches skills plus the full conversation history (FTS5 SQLite + LLM summarization) and brings in the relevant context.

Over weeks, this compounds. Repetitive work gets faster. The agent stops asking “how should I do X?” because it already has a working pattern for X, and keeps improving it.

Where to run Hermes Agent

Hermes Agent is designed to run persistently, which means you want a real server, not a laptop that goes to sleep. Two good options, depending on scale.

On a VPS

For individuals, small teams, or a single always-on assistant, a VPS is ideal:

  • Always on. Scheduled jobs and messaging-app integrations actually work.
  • Public, stable endpoint. A real IP that webhooks and gateways can reach without carrier-grade NAT in the way.
  • Low idle cost. Hermes Agent is light when idle. Most of the time it’s waiting for input or the next cron trigger.
  • Model-agnostic deployment. Use cloud model APIs on a small VPS, or scale up and run local Ollama models on the same machine.

Sizing starting points:

  • 2 GB RAM, 2 vCPUs: cloud APIs only
  • 8 GB RAM, 4 vCPUs: run Ollama with small local models alongside
  • 16 GB RAM, 6+ vCPUs: multiple concurrent agents or larger local models

On a dedicated server

If you’re running Hermes Agent for a team, an agency, or as part of a research/automation stack, a dedicated server gives you dramatically more room:

  • Large local models. Run 13B, 34B, or 70B models via Ollama without API fees, plus vector stores and supporting services on the same machine.
  • Multiple isolated agents. Use Hermes Agent profiles to run separate agents per client or project, each with its own skill library.
  • Heavy sandbox workloads. Docker, Singularity, and nested VMs for tool execution without squeezing a shared hypervisor.
  • GPU option. For serious local inference, a GPU-enabled dedicated server closes the cost gap with paid APIs on high-volume workloads.

Hermes Agent vs cloud agent platforms

Hosted agent platforms are convenient but come with trade-offs:

  • Per-seat or per-execution billing that scales with use
  • Conversation history and user data on someone else’s servers
  • Limited control over models, tools, and sandboxes
  • Risk of sudden account actions or price changes

Hermes Agent inverts all of those. You pay for the server and whatever model tokens you actually use. Your conversations, skill library, and user profile stay on your server. You pick the models. You pick the sandboxes.

Hermes Agent and OpenClaw

If you’ve used OpenClaw, Hermes Agent is worth a look because it fills one specific gap: the autonomous skill-learning loop. OpenClaw leans on human-authored skills and workspace integration; Hermes Agent autonomously builds its own skill library from successful completions.

The two aren’t mutually exclusive. Some people run both. OpenClaw handles broad ecosystem integration and quick one-off work, while Hermes Agent handles recurring workflows that benefit from compounding knowledge. Hermes even ships hermes claw migrate for bringing OpenClaw history over.

Real-world uses

From the community examples:

  • Personal assistant reachable via Telegram from your phone. Schedules, research, follow-ups, context across months.
  • Infrastructure ops. Scheduled health checks, log summaries, incident write-ups.
  • Research assistant. Paper ingestion, comparisons, persistent notes across weeks.
  • Agency workflows. Per-client isolated profiles, each with its own skill set.
  • Writing and coding pipelines. Novels, repos, long projects where old context still matters.

Getting started

  1. Deploy a VPS or dedicated server. Ubuntu 22.04/24.04 is the smoothest path.
  2. Run the Hermes Agent installer one-liner.
  3. Run hermes setup and pick a model provider.
  4. (Optional) hermes gateway to connect Telegram/Discord/Slack/WhatsApp.
  5. Use it daily and let the learning loop do its thing.

Step-by-step setup: Install Hermes Agent on a Linux server and Install Hermes Agent on Windows via WSL2.

Why this project is worth watching

Nous Research has been one of the more interesting open-weight model labs for years. Hermes 3, Hermes 4, and the Atropos RL framework all came out of the same group. Hermes Agent ships with research tooling (trajectory generation, compression, RL environments) that reflects that origin. It’s not just a consumer agent. It’s also a platform for evolving agents and generating training data.

Whether you run it as a personal assistant on a VPS or as part of a research stack on a dedicated server, the underlying design is the same: execute, evaluate, extract, refine, retrieve. That loop, running persistently on your own server, is why this project has grown so fast since launch.

Deploy a ServerPoint VPS or dedicated server and try Hermes Agent. For installation guides, see Hermes Agent on Linux or Hermes Agent on Windows.

← Back to all articles