15 Features to Unlock Hermes Agent
Leveraging Hermes Agent's persistent memory, context controls, multi-platform routing, and custom skills transforms it from a chatbot into a stateful operating system for automated workflows.
15 Features to Unlock Hermes Agent
TL;DR: You're leaving ~92% of Hermes Agent's capability on the table by treating it like a standard chatbot. By leveraging its persistent memory (SOUL.md), mid-flight context controls (/branch, /rollback), multi-platform routing, and custom skills, Hermes transitions from a simple conversational interface into a fully automated, stateful operating system for your workflows.
Most people install Hermes Agent and treat it like a smarter ChatGPT: wire up Telegram, pick a model, chat once, then close the tab.
You’re leaving ~92% of its power on the table.
Here are the 15 features that separate casual users from people who literally run their workflows on autopilot. Most long-time Hermes users have never touched half of these.
THE SETUP ALMOST EVERYONE SKIPS
1. SOUL.md + /personality
Hermes reads ~/.hermes/SOUL.md at boot. This single file becomes your agent’s permanent identity—tone, boundaries, audience, refusals—across every session and platform. Use /personality to switch modes mid-conversation. Edit once. Never type “you are a senior X expert” again.
2. MEMORY.md + USER.md
Persistent, searchable memory. MEMORY.md stores project truths; USER.md stores you (your role, preferences, trade-offs). Hermes pulls relevant context from weeks ago automatically. Stop re-explaining yourself every new chat.
3. /insights [days]
Instant analytics: token burn, project costs, what stalls the agent, what you reuse most. /insights 30 gives you a 30-day dashboard in one command.
4. /snapshot
One-command full state backup. Experiment wildly, then /snapshot restore <id> if it breaks. The “I’m about to refactor my SOUL.md” safety net.
MID-FLIGHT CONTROLS NOBODY USES
5. /branch (alias /fork) Git-style branching for conversations. Explore a risky path without losing your current context.
6. /rollback
Filesystem checkpoints. Agent made a destructive edit? /rollback restores any file (or the whole session) instantly.
7. /btw (alias for background) Ephemeral side question that uses full session context but doesn’t persist or burn tools. Perfect for quick gut checks like, "/btw what's the command to restart Docker?" without derailing your main coding session.
8. /steer + /queue Mid-task corrections without killing the run. Tell it “use staging, not prod” and the next tool call obeys—prompt cache stays warm.
9. /yolo, /fast, /reasoning Power toggles: skip approvals (carefully), force priority/low-latency mode, or adjust reasoning effort on o-series models.
THE PROVIDER LOCK-IN THAT DOESN’T EXIST
10. /model [–provider] [–global] Swap brains instantly: Claude Opus, GPT-5.5 via OAuth, OpenRouter, NVIDIA NIM, Kimi, Gemini, Bedrock—whatever you want. State carries over.
11. Auxiliary models Main brain on Opus, compression/summarization on Haiku, titles on a tiny model. Configure once; never overpay again.
THE REACH ALMOST NO ONE ACTIVATES
12. 17-platform gateway Telegram, Discord, Slack, WhatsApp, Signal, Email, SMS, Matrix, Mattermost, Feishu, WeCom, DingTalk, BlueBubbles, Home Assistant, QQBot + CLI + voice. One Hermes instance runs everywhere your team actually lives.
13. /voice (real-time on 4 platforms) Talk instead of type—in CLI, Telegram, Discord text, or live voice channels.
14. Cron + webhooks Native scheduler (“Every Friday at 5pm, summarize GitHub commits and post to Slack #standups”). Webhooks let external services (GitHub, Stripe, etc.) push straight to your DMs at zero LLM cost.
THE REAL UNLOCK
15. Skills = slash commands Skills aren't just text macros; they are packaged, repeatable workflows. Instead of manually typing out a complex, multi-step prompt every time you start a familiar task, Hermes lets you bundle instructions, specific tool permissions, and output formatting into a single command.
Hermes ships with built-in skills like /architecture-diagram (which automatically spins up graphing context) or /test-driven-development. More importantly, you can build your own. Spend 10 minutes defining your perfect workflow once, save it as a custom skill like /my-seo-analyzer or /sage (a custom daily content strategist), and it becomes callable forever across every platform. Type /, let it autocomplete, and watch it execute. You're building a personal software library, not just a list of prompts.