How to Use Hermes Agent in 2026: 16 Steps from Install to Automation

Get a server and deploy the agent

1. 1

   Pick a small KVM plan for the agent's server

   Hermes Agent runs around the clock, so give it a machine of its own. The walkthrough uses Hostinger's one-click Hermes Agent template and picks KVM 1 - the smallest plan, $5.84/mo at recording time with a promo (renewing at $11.99/mo) - and reports it runs one agent comfortably. Heavy all-day web browsing is the stated reason to step up to KVM 2, and you can upgrade later without reinstalling anything.

   

2. 2

   Set the admin login on the deploy form

   After checkout, the template only asks for two values. Keep ADMIN_USERNAME as hermes, click the eye icon on ADMIN_PASSWORD and copy the generated password into a password manager - everything visual from here needs it. The two optional fields at the bottom (Nexos and OxyLabs keys for their paid services) stay empty; this setup brings its own, cheaper model access instead.

   

3. 3

   Open the web dashboard and sign in

   When the container shows Running in Hostinger's Docker Manager, click Open on the Hermes Agent app. The sign-in page takes the username hermes and the password you saved - lowercase matters, as Tech With Tim's walkthrough of the same template points out. Lose the password later and Docker Manager's Environment tab shows and resets it.

   

Wire up the brain with OpenRouter

1. 4

   Meet the empty brain: no model configured

   First surprise: the agent is installed and running but cannot think. The chat page shows a Setup Required panel - 'Hermes needs a model provider before the TUI can start a session' - and the sidebar's error names the fix: run hermes model or set OPENROUTER_API_KEY. In the dashboard's rooms, Models is the brain, Cron the schedule and Skills its self-taught procedures; all three stay empty until a provider is wired, which is exactly the next step.

   

2. 5

   Create an OpenRouter key (bring your own brain)

   The walkthrough pays per use through OpenRouter: one account, hundreds of models. Sign up, copy the API key the moment the workspace page shows it - you will never see it again - then add $5 of credits, enough for weeks. While you are there, open API keys, edit the new key, name it Hermes Agent and set a $10 credit limit that resets weekly; that cap is your spend guardrail.

   

3. 6

   Paste the key into the dashboard's Keys page

   Back in the dashboard, open Keys and expand the OpenRouter row. Click Set next to OPENROUTER_API_KEY, paste the sk-or key and save. This page also holds OAuth logins (Nous Portal, ChatGPT, Claude, Gemini), so it is the one place to revisit whenever a provider needs wiring. Then go to Models, hit Change on Main model, pick DeepSeek V4 Flash through OpenRouter and confirm the reload.

   

4. 7

   Say hello to a thinking agent

   Type hello into the chat. The boot panel lists what just came alive - 21 toolsets including browser use, code execution, cron jobs, delegation and memory, plus 77 skills in 12 categories at recording time - and the MODEL badge flips to deepseek-v4-flash, Live. If a send ever errors, start a New chat and retry; the walkthrough's first hiccup cleared on the second attempt.

   

Reach it from your phone

1. 8

   Pair Telegram with one QR scan

   Channels is where the agent gets a phone number. Telegram's Quick setup is the recommended route: click CREATE WITH QR, scan with your phone and confirm - Hermes Agent creates the bot and detects your Telegram user ID automatically. After the scan, click save and restart and wait for the gateway to flip back to Running, send any message to test, then tap /sethome so scheduled-job reports land in that chat. The QR expires in about three minutes, so keep the phone in hand.

   

2. 9

   Or bring your own bot via BotFather

   Prefer a bot under your own name? Tech With Tim's course shows the manual route: message @BotFather, send /newbot, choose a display name and a username ending in bot, then paste the token into Channels > MANUAL SETUP. The same flow asks for allowed user IDs - the numeric ID Telegram assigns you - so only trusted accounts can ever talk to your agent.

   

Memory, skills and the desktop app

1. 10

   Point the desktop app at your server

   Install the desktop app from hermes-agent.nousresearch.com; when its onboarding offers a provider, choose 'I'll choose a provider later' so you do not spawn a second local agent. Open Settings > Gateway, select Remote gateway and paste the server address - the dashboard URL up to the first slash, highlighted in the address bar here. Sign in with hermes plus your password, save and reconnect: the app reloads showing your server sessions, Telegram chats included. A quick 'what was my first message to you?' proves the memory followed.

   

2. 11

   Correct it once - the fix sticks forever

   This is the moat. When a draft post came back too wordy, the walkthrough answered in plain language - 'short, punchy sentences, three paragraphs max, no headings or bullets. Save that as a preference' - and the Saved to memory chip fired on screen. A brand-new session (Cmd+N) then wrote the next post in exactly that style with zero context carried over. Memory, skills and the SOUL.md persona all live in files on your server, not in someone else's account.

   

3. 12

   Read what it knows on the Capabilities page

   Capabilities lists every skill (77 at recording time), toolset and MCP connection; click any row to read its plain-text source - the hermes-agent skill even documents its own MIT license and GitHub repo. Skills marked as learned are ones it wrote itself while working. The Skills Hub at the bottom adds searchable community procedures; Tech With Tim counts about 90 built in with hundreds more installable.

   

Automation, sub-agents and the cost bill

1. 13

   Let it schedule its own watch jobs

   The signature trick costs one message: 'watch YouTube for AI productivity trends, build yourself a reusable skill for the check, and message me only when there is something new.' The agent installed a dependency, wrote the skill and created the ai-trend-watch job shown here, running every 240 minutes with its full prompt readable. Jobs can be paused or triggered by hand, and Blueprints ship templates from morning briefings to price watches.

   

2. 14

   Read the real bill on OpenRouter Activity

   The scariest-looking page is the friendliest: a stretch of real agent work - 105 requests - billed $0.13. The volume is the surprise: nearly 4 million tokens, because the agent resends its instructions, memory, tool list and chat history every single turn. Providers cache the unchanged part, which is why the 75.1% cache hit rate matters more than the token count. At that rate the opening $5 covers roughly 4,000 messages on DeepSeek V4 Flash, per the walkthrough's own math.

   

3. 15

   Fan out big jobs to sub-agents

   For work that splits, say so: 'Plan me three days in Lisbon - one sub-agent on neighborhoods, one on food, one on a day trip, one on getting around.' The trace shows four DeepSeek V4 Flash delegations running in parallel under a 4 Subagents chip before the main agent merges one itinerary. Parallel helpers are why the walkthrough calls Hermes Agent faster than single-threaded agents on research jobs.

   

4. 16

   Split work and life with profiles

   Behind the channel grid sits the multi-agent move: New profile creates an independent Hermes Agent environment with its own config, skills, SOUL.md and connectors, cloned from default if you want a running start. A work agent and a personal agent stop leaking context into each other. Coming from OpenClaw, just ask the agent to import your OpenClaw configuration and it walks the migration with you.

   

Frequently asked questions