curl | bash без ручного Python → hermes doctor (~12 checks) → hermes setup для LLM → hermes gateway install для Telegram 24/7. Ниже — diff с OpenClaw, env matrix, six-step runbook, BotFather wiring, error table и когда аренда Mac Mini M4 bare metal оправдана для skill compounding и launchd uptime.
Зачем ставить Hermes Agent в 2026 — и чем он отличается от OpenClaw
В open-source agent tutorials пересекаются Hermes Agent, OpenClaw, LangGraph и CrewAI. Разделение простое: Hermes — personal on-call operator на вашей машине — USER.md, MEMORY.md и skills в ~/.hermes, Gateway для каналов. OpenClaw — multi-channel hub + ClawHub skills. LangGraph/CrewAI — orchestration frameworks: память и deploy собираете сами.
Self-improving skills: Repeat tasks грузятся из ~/.hermes/skills/. Community reports: skill count с single digits до ~12 за 30 дней, token на similar jobs −~38% — efficiency signal, не quality score.
Model-agnostic: Anthropic, OpenAI, OpenRouter, GitHub Copilot, Ollama — keys в ~/.hermes/.env.
MCP extensions: Model Context Protocol servers — GitHub, browser automation и др.
Install pain point: Пройти CLI once ≠ цель — нужен Gateway 24/7. Host compare: three-way matrix; memory compounding: three-layer guide.
Требования Hermes Agent: macOS, Linux или VPS?
| Параметр | Minimum | Recommended (production Gateway) |
|---|---|---|
| OS | macOS 12+, Ubuntu 20.04+, WSL2 | macOS 14+ (Apple Silicon) или Ubuntu 24.04 LTS |
| Python / Node | Script ставит 3.11 / Node 18+ | Default Node 22, без sudo |
| RAM | 4 GB (только API + Gateway) | 16 GB (browser automation + local 7B fallback) |
| Disk | ~1.5 GB (skill cache incl.) | ≥ 20 GB SSD |
| Network | GitHub raw + LLM API reachable | Stable low-latency egress |
| Host scenario | 24/7 uptime feel | Best for |
|---|---|---|
| Laptop lid closed | ≈ 60%, skill compounding breaks | POC only |
| x86 VPS 4 GB | ≈ 99.5%, no macOS skill paths | Linux-only users |
| Rented Mac Mini M4 | Datacenter SLA, native launchd | Telegram production, Camoufox |
Установка Hermes Agent: runbook из шести шагов (macOS / Linux)
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash source ~/.zshrc
Headless: bash -s -- --skip-browser. Script ставит uv, Python 3.11, Node 22, ripgrep, ffmpeg, clone в ~/.hermes/hermes-agent/, hermes в PATH. Success line: ✅ Hermes Agent installed at ~/.hermes.
Run install script: curl выше. Windows — install.ps1, но production favors WSL2 + Linux script.
Reload shell: source ~/.zshrc или source ~/.bashrc — fix hermes: command not found.
Health check: hermes doctor — Python, Node, ripgrep, ffmpeg, PATH, API keys (~12 checks). hermes --version → v0.14.x или v0.15 candidate.
Configure LLM: hermes setup wizard или hermes model для OpenRouter / Anthropic / Ollama. Keys в ~/.hermes/.env, напр. hermes config set OPENROUTER_API_KEY sk-or-xxxx.
CLI smoke test: hermes, «summarize today's AI news» или «count files in directory» — confirm tool calls и model routing.
Upgrade & migrate: hermes update; после upgrade — hermes config check && hermes config migrate.
Hermes Agent Telegram: от BotFather до gateway install
CLI alone = local chat. Для tasking с телефона нужен Gateway. Flow: @BotFather → /newbot → save token → @userinfobot для numeric user ID.
hermes gateway setup hermes gateway install hermes gateway start hermes gateway status
Wizard пишет TELEGRAM_BOT_TOKEN и TELEGRAM_ALLOWED_USERS. Linux production: sudo hermes gateway install --system + journalctl -u hermes-gateway -f; macOS — launchd user service.
| Error / symptom | Cause | Fix |
|---|---|---|
hermes: command not found | PATH not loaded | Reopen terminal or source rc |
API key not set | No LLM configured | hermes model |
| Group chat silent | Bot privacy mode | Disable Group Privacy in BotFather, kick and re-add |
ModuleNotFoundError: dotenv | System Python used | ~/.hermes/hermes-agent/venv/bin/hermes |
Tip: Also useful: hermes tools, hermes config check. Declare mcp_servers in ~/.hermes/config.yaml; headless — hermes mcp login + SSH tunnel for OAuth.
После install: skill compounding и hard data Mac Mini M4
Star scale: Июнь 2026 — hermes-agent ~174 720 stars. Install и Telegram docs fastest iterate на macOS path.
Power draw: M4 idle 4–6 W; Gateway + automation typically 15–25 W — ok для always-on.
Unified memory: 16 GB UMA — Gateway residency ~4 GB class, shared address space с Camoufox и SQLite WAL — меньше swap thrash vs small x86 VPS.
Closed laptop, Raspberry Pi I/O limits, Linux VPS без macOS browser skills — не install failures, а wrong host → Gateway SLA break. Для Telegram 24/7 с self-writing skills — MESHLAUNCH Mac Mini M4 bare metal rental: same curl over SSH, launchd supervision, tarball ~/.hermes before offboarding. Цены аренды; ops boundaries — центр помощи.
Нет. install.sh через uv ставит Python 3.11 и Node 22. Если doctor fails — вызывайте venv hermes, не system Python.
После отключения Group Privacy в BotFather удалите бота из группы и пригласите снова. Проверьте TELEGRAM_ALLOWED_USERS. Можно арендовать M4 для изолированного Gateway-теста.
Disk memory persists, но offline Gateway stops skill compounding и Cron. Production: hermes gateway install на 24/7 host — см. memory architecture guide.