2026 Hermes Agent install walkthrough: One command to Telegram 24/7 uptime

Hermes Agent took off after its February 2026 launch with community reports that tasks auto-write Skills and similar jobs later cut tokens by ~38%. Unlike OpenClaw (multi-agent workflows) or LangGraph (graph state machines), Hermes stands out with native cross-session memory, 20+ messaging Gateway surfaces, and MCP plug-and-play. Many people star the repo then stall on three steps:

• Fragmented environment: assuming you must pre-install Python 3.11, Node 22, ripgrep, ffmpeg—the official script installs via uv, but an unrefreshed PATH yields hermes: command not found.
• No model configured: install succeeds yet API key not set appears; CLI opens but cannot reason.
• Gateway not always on: Telegram Bot configured, laptop lid closed, Bot “dead,” Skill compounding and cross-week memory stop.

In one line: Hermes goes smoothly when you treat install + model + always-on gateway as one pipeline, not three separate experiments.

The stable line is v0.15.1 (2026.5.29), fixing v0.15.0 Dashboard reload loops on loopback (see official release). Commands below target macOS / Linux / WSL2; native Windows PowerShell can use official install.ps1; Dashboard chat still favors WSL2.

The install script’s only hard prerequisite is Git (git --version must work). Python, Node, ripgrep, and ffmpeg are detected and installed by the script—no sudo required.

On headless servers (no GUI browser automation) add --skip-browser to avoid Camoufox-related deps. Config lives in ~/.hermes/.env and ~/.hermes/config.yaml.

On macOS, Linux, or WSL2 run the official script (matches install docs):

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

# headless VPS optional
# curl -fsSL ... | bash -s -- --skip-browser

source ~/.zshrc
hermes doctor
hermes --version
hermes setup
hermes

The script installs uv → Python 3.11 → Node.js 22 → ripgrep / ffmpeg → clones to ~/.hermes/hermes-agent/ → adds hermes to PATH → pulls default Skills registry. Success looks like Hermes Agent installed at ~/.hermes in the terminal.

1. Run curl install: ensure Git is available; set https_proxy first behind corporate proxy.
2. Refresh shell: source ~/.zshrc or source ~/.bashrc, else command not found on the next line.
3. Health check: hermes doctor checks ~12 items (Python, Node, PATH, API key, etc.).
4. Confirm version: hermes --version ideally ≥ v0.14.x; upgrade with hermes update.
5. Configure model: hermes setup or hermes model writes ~/.hermes/.env.
6. Smoke chat: run hermes, ask to count files in the current directory or summarize today’s AI news to verify tools.

Short link: curl -fsSL https://get.hermes-agent.org | bash (community docs). Windows PowerShell: iex (irm .../install.ps1); Android Termux uses the Linux script with official Termux notes.

To reach your agent from a phone, complete Bot creation + Gateway config + system service in three steps.

1. Create Bot: in Telegram find @BotFather, /newbot, pick a name; username must end with bot; save the token.
2. Get user ID: message @userinfobot for a numeric ID for allowlisting.
3. Interactive setup (recommended): hermes gateway setup, choose Telegram, paste token and ID.
4. Or manual .env: TELEGRAM_BOT_TOKEN=... and TELEGRAM_ALLOWED_USERS=yourID.
5. Foreground test: hermes gateway, message the Bot from your phone and confirm replies.
6. Background service: on macOS hermes gateway install + hermes gateway start (launchd); on Linux sudo hermes gateway install --system + systemd, logs journalctl -u hermes-gateway -f.

Groups: default privacy mode only responds to / commands. For group use disable Group Privacy in BotFather and re-invite the Bot after removing it once. Discord / Slack use the same hermes gateway setup; see official messaging docs.

hermes gateway install
hermes gateway start
hermes gateway status

# Linux boot autostart (optional)
sudo hermes gateway install --system
sudo hermes gateway start --system

• GitHub scale (2026-06): ~175,000 stars, MIT license, 295+ contributors (repo).
• Install size: ~1.5 GB with Skills cache; ~/.hermes/ often grows 200–600 MB over three months (Skills and logs).
• Skill compounding: community reports Skills 3 → 19 in 30 days with ~38% token drop on similar tasks—requires host 24/7 online or compounding stops.
• Gateway RAM (macOS summary): Gateway + SQLite alone ~1.2–1.8 GB; with Ollama 8B plan 24 GB unified memory.

Optional: declare mcp_servers in ~/.hermes/config.yaml for GitHub, databases, etc.; on GUI-less servers use hermes mcp login + SSH tunnel for OAuth.

Common laptop and oversubscribed VPS gaps: ① unstable gateway long connections (sleep, NAT, neighbor IO); ② macOS browser-automation Skills do not 1:1 on pure Linux; ③ without dedicated RAM parallel sub-agents OOM while hermes doctor still passes.

Safer path: rent bare-metal Mac Mini M4 monthly on JEXCLOUD, SSH the same install commands as this article, launchd KeepAlive for Gateway, ~/.hermes always on your disk—no virtualization oversubscription, stable Telegram inbound, rsync Skills and memory before return. Specs and M4.S / M4.M / M4.XL sizing on pricing; order and region latency on order and help center.