Build Your First AI Agent — Free

1GET A SERVER (~$10/MO)

1. Grab a Hostinger KVM 2 VPS (2 vCPU / 8GB is the sweet spot for one agent) — hostinger.com/vps-hosting
2. OS template: pick Ubuntu 24.04 (plain is perfect — Hermes installs itself in step 3)
3. Set a strong root password, note your server IP

2GIVE IT A BRAIN (PICK A MODEL)

Your agent needs an AI model to think with. Here are three clean ways to power it — pick one:

1. Kimi K3 — the easy button (my pick)
   Frontier reasoning + vision on a flat monthly subscription that includes a ready-to-use coding key — no per-token bills, no billing surprises. Subscribe at kimi.com, copy your coding API key from your account, and point Hermes at https://api.kimi.com/coding/v1 (Anthropic-compatible). That’s it — done.
2. Already pay for ChatGPT? Use Codex — no API bill
   OpenAI’s Codex signs in with your existing ChatGPT Plus/Pro account and runs GPT for you — no API key, no per-token charge. Your subscription does the work. Install it and log in: brew install codex # Mac (or: npm i -g @openai/codex) codex login # sign in with your ChatGPT account Then bridge Codex into Hermes as an OpenAI-compatible model (we use auth2api for this). This bridge is the one genuinely fiddly part of the whole build — if it fights you, that’s exactly what our free done-for-you support is for.
3. Want a raw OpenAI or Claude key instead?
   Create one at platform.openai.com or console.anthropic.com, add a payment method, and paste it into hermes model. Heads-up: this is pay-as-you-go per token, billed separately from any chat subscription — fine for light use, keep an eye on it for heavy use.

One more key — your Telegram bot (this is how you’ll actually talk to your agent):

4. Open Telegram → message @BotFather → send /newbot → give it a name → copy the bot token it hands back. Keep it somewhere safe — you’ll paste it into Hermes in the next step.

3INSTALL HERMES — THE BEST ALL-AROUND HARNESS

Why Hermes: it's the only agent with a built-in learning loop — it writes its own skills from experience, remembers you across sessions, and lives in Telegram, WhatsApp, Discord, Slack, Signal and your terminal from one gateway. Any model, no lock-in.

SSH into your server (Mac/Linux: Terminal · Windows: PowerShell) — swap in your server IP:

THEN PASTE THE ONE-LINE INSTALL

On Windows? Skip the SSH box and paste this into PowerShell instead:

The installer handles everything — Python, Node, browser tools — then drops you into a setup wizard. When it asks for your model, paste the key from step 2, or take the shortcut below.

SHORTCUT — ONE LOGIN, EVERY TOOL

One sign-in wires up 300+ models plus web search, image generation and a cloud browser under a single subscription — no juggling separate keys. Then connect Telegram (paste your bot token from step 2) and you're live. It installs as a service that survives reboots.

4SAY HELLO + GIVE IT A JOB

1. Open your bot in Telegram, send "hello" — it should answer as your agent
2. Give it ONE job first: "Every morning at 8, check [X] and message me a summary"
3. Add skills/memory gradually — one working job beats ten broken ones

STUCK? WE'VE GOT YOU

If you're 30 minutes in and it's still not answering — that's not a you-problem, it's a config rabbit hole (usually the Codex bridge or the model key). Don't grind on it. We do these all day and we're happy to finish yours with you, live, in about 15 minutes.