.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
This is the hub page for everything we've written about Hermes Agent, the open-source self-hosted AI agent that Nous Research shipped in February 2026. If you've been bouncing between tabs trying to figure out where to start, this guide walks the whole arc in the order most people end up taking. Decide if Hermes is the right pick. Install it. Migrate your old setup if relevant. Wire up a messaging channel. Harden the box for production. Learn the memory model so you can debug the agent. Recover when something inevitably breaks.
Read it top to bottom on a first pass, then bookmark the sections you need to come back to. Every numbered topic links out to a longer tutorial covering it in detail.
1. Decide if Hermes is the right agent for you
Hermes is one of three popular self-hosted agents we host on LumaDock. They're not interchangeable. The choice depends on how much polish you want, how much control you need over memory and how comfortable you are with a fast release cadence.
Start with the analytical comparison in Hermes Agent vs OpenClaw for self-hosting. It walks the operational trade-offs honestly. OpenClaw has a bigger skill library and a longer track record on every messaging platform. Hermes has a cleaner memory model, an explicit learning loop and a faster release pace. Neither one is uniformly better.
If you're already running OpenClaw and just want to know if a switch is worth your weekend, the decision framework in when to migrate from OpenClaw to Hermes Agent covers the four most common reasons people make the jump and the two reasons to stay put.
If your use case is more "I want a simple, predictable bot that doesn't try to learn behaviours on its own", ZeroClaw is worth a look before either of the heavier options. The ZeroClaw vs OpenClaw comparison sketches where the lightweight option wins. If you decide to go that route, our ZeroClaw VPS setup guide picks up from there.
For the broader OpenClaw library (templates, channels, security writeups, cost guides), the OpenClaw complete guide is the equivalent of this article on the OpenClaw side.
2. Get Hermes installed
Once you've picked Hermes, the next decision is install path. The install Hermes Agent on Ubuntu 24.04 walkthrough covers the 90-second curl-pipe-bash flow plus the manual fallback for the few edge cases where the official installer chokes. Read the first half if you just want the agent on the box. Read the second half if your environment has anything unusual about it: locked-down ports, non-standard Python, air-gapped servers.
If you'd rather skip the install entirely, the LumaDock Hermes Agent template ships with the agent pre-installed on Ubuntu 24.04, dashboard ready to pair over SSH and the hermes CLI on the PATH. The Hermes Agent VPS setup guide in our knowledge base walks the first-boot menu screen by screen.
3. Migrate from OpenClaw (only if you were running OpenClaw)
If you're starting from scratch, skip this section. If your previous agent was OpenClaw, the canonical walkthrough is how to migrate from OpenClaw to Hermes Agent. It covers the hermes claw migrate command end to end including the pre-flight checks and the post-migration verification.
The migration is rarely a perfectly clean run. When it isn't, our fix hermes claw migrate failures and silent skips article catalogues the eight failure modes we've seen most often. Silent skill skips. Persona conflicts. Dirty workspaces. Missing API keys. Schema mismatches. Partial state.db transfers. Broken symlinks. The friendly-but-confusing dry-run output.
For people who'd rather not commit to a switch immediately, our run OpenClaw and Hermes Agent on the same VPS guide walks the coexistence pattern. Separate users, separate ports, separate systemd units, no shared state. You get to trial Hermes for a few weeks with your old agent still answering on the channels it owned.
4. Pick a messaging channel
Hermes only becomes useful once it's reachable from somewhere you read messages. The most common channels each get a dedicated tutorial.
Hermes Agent Telegram bot setup and reliability is the one most people start with. Telegram is the cheapest channel to get right: free BotFather, polling out of the box, no inbound port exposure. The article covers the allowlist pattern that keeps strangers from running up your token bill.
For Discord servers, Slack workspaces and yes Telegram in parallel, our multi-platform messaging gateway for Hermes Agent walks the multi-channel setup with shared memory and identity mapping. This is the "one brain, many faces" article.
WhatsApp is the polarising one. The Baileys-based bridge works but it's a third-party reverse-engineered library and accounts do get banned. Our Hermes Agent WhatsApp setup with Baileys tutorial covers the dedicated-number pattern that minimises ban risk plus the voice-note transcription flow people love.
For overnight email triage and background tasks, Hermes Agent email gateway with IMAP and SMTP shows the Gmail app-password setup, the inbox-filter pattern that keeps Hermes from answering marketing emails and the outbound-only configuration people use for transactional notifications.
5. Make it production-grade
Once Hermes is on a channel, it needs to stay up. Three articles cover the operational baseline.
The simplest path is run Hermes Agent 24/7 with systemd. A unit file, a restart policy and a journald log rotation get you to "the agent recovers from crashes by itself" in about fifteen minutes. The article includes a hardened unit (NoNewPrivileges, ProtectSystem, ProtectHome) for people who want defence in depth without reaching for containers.
If you prefer containers, Hermes Agent on Docker Compose with persistent state covers a production-ready Compose file with healthchecks, named volumes for state.db and skills plus a watchtower service for auto-updates. The persistent-state section is the one most people get wrong on the first try.
For anything exposed to the public internet (the web dashboard, a webhook receiver, the community WebUI), the Hermes Agent behind Nginx with HTTPS and basic auth guide is what you want. Let's Encrypt, basic auth, WebSocket passthrough and a sensible default location block.
Two more articles round out the production set. The Hermes Agent production hardening checklist is a 15-item walkthrough covering non-root operation, command-approval mode, sandboxing of shell tools, secrets management with sops or systemd-creds and the audit log everyone forgets to enable. The Hermes Agent backups for state.db, skills and identity article covers the SQLite-aware backup script, GPG encryption at rest, the systemd timer schedule that runs it nightly and how to restore on a different host without breaking your memory.
6. Understand the agent's brain
Hermes is more debuggable than most agents because its memory is flat files plus one SQLite database. You can read it with a text editor and grep it with normal tools. The Hermes Agent memory: SOUL.md, MEMORY.md and state.db article maps the three layers (persona, user facts, session state) and shows the hand-editing patterns we use to fix common memory bugs without nuking the agent's history.
Skills are the other half of the brain. The agent writes its own skill files as it learns, and the edit Hermes Agent skills without breaking the loop tutorial covers the user_locked: true frontmatter that protects your hand-edited skills from being overwritten, the fork pattern for adapting community skills and the git-based workflow for sharing skills across boxes.
Cost is the third concern most people raise after a week of use. Cut your Hermes Agent token bill in half breaks down the issue #4379 overhead numbers (the ~13.9K tokens of fixed context per call), the four provider-routing tricks that move spend from Anthropic to cheaper open-weight models for the right kinds of tasks, the /compress command for long conversations and when self-hosting an open-weight model on a GPU box pays for itself.
7. Recover when things break and reach the agent remotely
The two articles in this final group are the ones you'll bookmark and forget about until you need them.
The most common operational failure in Hermes is the SQLite locking pattern documented in issue #3139. Our fix the SQLite database is locked error in Hermes walkthrough covers the WAL-mode tuning that prevents most of these locks, the busy_timeout and mmap_size settings worth changing, the recovery steps when a lock has wedged a session and the diagnostic commands that tell you if session_search has been silently disabled.
For remote access from your laptop or phone, the community WebUI is the option most people want. Hermes WebUI for remote access to your agent covers the install, the OIDC authentication setup so it isn't open to the public internet and the Nginx configuration that puts it behind the same reverse proxy as everything else on your VPS.
Where the LumaDock template fits in
Everything above works on any VPS. If you don't already have one or you'd rather not deal with the install, the LumaDock Hermes Agent VPS template is the route most readers take. Pick a plan. The server boots with Hermes pre-installed on Ubuntu 24.04. You SSH in. The first-login menu walks you to either the web dashboard or the terminal wizard. You configure your LLM provider and your first channel in a browser. You send the bot a message. The whole loop is under five minutes on a normal connection.
If you're switching from OpenClaw, the same template includes the migration tool. The path is: order the template, tarball your old ~/.openclaw to the new box, run hermes claw migrate, verify, point your channels at the new agent. Our Hermes Agent VPS setup guide in the knowledge base covers the screen-by-screen flow for the first boot.
