@shmidtqq: You hooked up Telegram. You pi...
You hooked up Telegram. You picked a model. You type prompts, get answers, close the tab.
You're using 8% of Hermes.
The other 92%, persistent memory, session branching, file rollbacks, voice mode, 17-platform reach, custom slash commands, sits there untouched while you treat a fully-loaded agent like a slightly smarter ChatGPT.
This article is the 92%. 15 features, ranked by impact. Most Hermes users haven't touched a single one.
Quick context for newcomers: Hermes is a provider-agnostic AI agent. 100+ pre-built skills. Runs on 17 messengers from one process. Swaps models mid-session without restarting. Built portable, extensible, yours, not locked to one vendor like every other agent on the market.
Now the list.
PART 1: The Setup You Skipped
1. /personality + SOUL.md
Hermes reads SOUL.md at boot. Whatever's in it becomes your agent's voice forever, every session, every platform. /personality switches between named personas on the fly.
Stop typing "you are a senior X expert" at the start of every chat. Write it once.
2. MEMORY.md + USER.md
Two persistent files, read every session. MEMORY.md = project notebook. USER.md = what it knows about you. Indexed with FTS5 + LLM summarizer, so a memory from 8 weeks ago surfaces in today's session.
You stop re-explaining yourself.
3. /insights [days]
Cross-session analytics. Tokens burned, providers used, where the agent stalled, what you keep returning to. /insights 30 = last month at a glance.
4. /snapshot
Save full Hermes state before doing something risky. Break things on purpose. /snapshot restore
PART 2: Mid-Flight Controls
5. /branch (alias /fork)
Branch the session like a git commit. Try a riskier path without burning your good context. Doesn't pan out? Come back.
6. /rollback
Filesystem checkpoints. Agent nuked your code? Skip git /rollback. Hermes saves every file it touched.
7. /btw
Ephemeral side question. Uses session context, calls no tools, doesn't get persisted. The "quick gut check" command.
8. /steer and /queue
Three tool calls into a long run, you realize the agent is hitting prod instead of staging. Don't kill it. /steer use staging not prod. Next tool call sees the note. Cache stays warm. /queue lines up the next turn without breaking the current one.
9. /yolo, /fast, /reasoning
Three power toggles. /yolo skips dangerous-command approvals. /fast flips to OpenAI Priority or Anthropic Fast Mode. /reasoning sets effort level for reasoning models. Most people stay on defaults forever and wonder why their sessions feel slow.
PART 3: The Provider Lock-In That Isn't
10. /model [--provider] [--global]
One command swaps the model. No restart. Anthropic Opus 4.7, OpenAI Codex (GPT-5.5 via OAuth, no API key), OpenRouter, NVIDIA NIM, Kimi, Gemini, AWS Bedrock, Vercel AI Gateway, Xiaomi MiMo, Step Plan, Arcee.
/model anthropic:claude-opus-4-7 for the heavy lift. /model openrouter:kimi-k2.6 for grunt work. State carries over.
11. Auxiliary Models
The agent compresses context, summarizes sessions, generates titles, runs vision. You can route each to a different model. Opus 4.7 for the main brain, Haiku 4.5 for compression, a tiny model for titles.
Stop paying Opus rates for Haiku-grade work.
PART 4: The Reach You Never Activated
12. The 17-Platform Gateway
Telegram, Discord, Slack, WhatsApp, Signal, Email, SMS, Matrix, Mattermost, Feishu, WeCom, DingTalk, BlueBubbles, Home Assistant, QQBot, plus CLI and voice. One Hermes process drives all of them.
13. /voice: Real-Time Voice on 4 Platforms
CLI, Telegram DMs, Discord channels, Discord voice rooms. Type /voice and talk. Walking, driving, eyes on something else, done.
14. Cron + /webhook-subscriptions
Built-in scheduler. Plain language schedules.
"Every Friday 5pm, summarize this week's GitHub commits, post to Slack #standups."
Pair with /webhook-subscriptions for the inverse: GitHub, Vercel, Stripe, uptime checks push payloads straight to your DMs. Zero tokens. Zero LLM cost. Zero latency.
You're paying Zapier for this every month.
PART 5: What Separates Real Users from Tourists
15. Skills Are Slash Commands
This is the one. 100+ skills out of the box, every one a slash command. Hit /, autocomplete fires.
```
/architecture-diagram, /excalidraw, /manim-video, /research-paper-writing, /linear, /google-workspace, /imessage, /youtube-content, /codex, /claude-code, /test-driven-development, /systematic-debugging.
```You can write your own. I built /sage. It spots outliers in my niche, scouts trends, drafts QTs and threads in my voice. Built once. Type /sage in any session, any platform, runs forever.
Tourists use slash commands once a week. Real users built their entire workflow into them.
You paid for an agent with persistent memory, 100+ skills, filesystem rollback, session branching, mid-flight steering, 17-platform reach, real-time voice, multi-provider routing, auxiliary model routing, cron, webhooks, and custom slash commands.
You used it as a slightly fancier telegram bot.
The tool wasn't underdelivering. You never gave it the instructions it was waiting for.
I publish settings, configurations, and unreleased Hermes tricks like these on my Telegram channel before they appear elsewhere.
If you want to get the next 15 and the SOUL.md, USER.md, and /sage files I run daily, subscribe to my tg: https://t.me/+JmDeelv5UCwwMTcy
That's where you'll find all the interesting stuff!