Last updated on

Hermes v0.18 Command Panorama: 82 Slash Commands Across 12 Real-World Scenarios

If you are new to Hermes Agent, the sheer number of slash commands can be overwhelming: /new, /model, /compress, /background, /rollback… Which ones are daily drivers, and which are niche tools for special situations?

The good news is that Hermes v0.18 organizes them clearly. The central registry at hermes_cli/commands.py (COMMAND_REGISTRY) currently lists 82 slash commands, plus 38 CLI subcommands in hermes_cli/subcommands/, bringing the total to over 120.

This article doesn’t just dump a list. It organizes them into 12 real-world scenarios, so you go from “I know /help” to “I can command this thing blindfolded.”

1. Architecture Overview

Commands live on two dimensions:

  1. Slash commands (/command) — typed inside a conversation session; affect the agent’s current state.
  2. CLI subcommands (hermes <command>) — executed from the terminal; manage configuration, installation, and deployment.

Each slash command is defined as a CommandDef dataclass in hermes_cli/commands.py, with name, description, category, aliases, and argument hints. CLI subcommands live in individual files under hermes_cli/subcommands/.

Think of it this way:

  • Slash commands handle “how to use Hermes while chatting” — switching models, compression, status checks.
  • CLI subcommands handle “how to set up and maintain Hermes” — installing tools, configuring gateways, scheduling cron jobs.

Scenario 1: Session Management — Conversation Basics

The most-used set of commands. Learn how to start, undo, retry, and save conversations.

/new [name] (alias /reset)

Start a fresh session. No arguments clears history immediately. Pass a name to set a session title on creation.

> /new
Session reset. You can start a new conversation.
> /new Email Integration Research
Started a new session titled "Email Integration Research"

/retry

Re-send the last user message to the agent. Useful when the response was unsatisfactory.

> Write a Docker Compose for PostgreSQL
(... not quite what you wanted ...)
> /retry
(Re-sends the same message; agent tries again)

/undo [N]

Undo N conversation turns. Defaults to 1. Undo 3 with /undo 3.

> /undo 3
Undone the last 3 messages

/title [name]

Name the current session. Without arguments, shows the current title.

> /title Backend API Performance Optimization
Session title updated to "Backend API Performance Optimization"

/clear (CLI only)

Clear the terminal and start a new session.

/history (CLI only)

View the full conversation log.

/save (CLI only)

Save the current conversation to a local file.

Scenario 2: Session Search & Resume — Pick Up Where You Left Off

All Hermes conversations are persisted in SQLite with full-text search, so you can search and resume from any platform.

/sessions

Browse all historical sessions in reverse chronological order.

> /sessions
47 sessions found:
1. Backend API Performance Optimization (Jul 19)
2. Email Integration Research (Jul 18)
3. Hermes Deployment Guide (Jul 17)
...

/resume <id | name>

Restore a specific session’s context. Supports numeric ID (/resume 3) or title keyword (/resume deployment).

> /resume 3
Resumed session "Hermes Deployment Guide". Context loaded.

/restart

Restart the current session (keep the session ID, clear the context).

Under the hood

Hermes’ session system uses ~/.hermes/state.db (SQLite + FTS5 full-text index) storing: session ID, platform source, model config, system prompt snapshot, full message history, token counts, and timestamps.

Scenario 3: Model Switching & Control — Swap LLM Providers Instantly

/model [name] [--provider <p>] [--session | --global]

One of the most powerful commands. Supports:

  • /model — interactive model picker
  • /model gpt-4o — switch to a specific model
  • /model --provider deepseek — switch to a provider
  • /model --session — per-session only
  • /model --global — persist globally
> /model anthropic/claude-sonnet-4
Model switched to anthropic/claude-sonnet-4 (persisted)
> /model --session qwen-max
Using qwen-max for this session only

/fast

Toggle fast mode. On OpenAI this enables Priority Processing; on Anthropic it switches between Normal and Fast mode.

/yolo

Toggle YOLO mode. When enabled, Hermes skips all dangerous-command confirmations.

> /yolo
YOLO mode enabled! (Use with caution)

/reasoning

Control model reasoning depth (works with o1, o3, and other reasoning-capable models).

/codex-runtime

Switch between local terminal and remote Codex runtime environments.

Scenario 4: Context Compression & Optimization — Keep Chatting Without Hitting Limits

Long conversations burn tokens. Hermes provides a multi-level compression system.

/compress [here [N] | focus <topic> | --preview] (alias /compact)

Three compression modes:

# 1. Default: keep last 10 turns, compress the rest
> /compress

# 2. Keep last N turns
> /compress here 15

# 3. Focus on a specific topic
> /compress focus performance

# 4. Preview without executing
> /compress --preview

/status

One-stop view of current session state — model, tokens, context usage.

> /status
Model: anthropic/claude-sonnet-4
Tokens: 12,847 input / 3,291 output
Context: 76% (13K / 17K)
Session: Backend API Perf (ID: abc-123)
Mode: non-YOLO

/usage (requires Nous Portal)

View account usage and quotas.

/insights

Analyze conversation patterns — token trends, turn counts, and more.

Scenario 5: Safety & Rollback — Never Fear Destructive Agent Actions Again

This is one of Hermes’ biggest differentiators in the AI agent space: automatic filesystem checkpointing and rollback.

/rollback [number]

Restore the working directory to a checkpoint. Hermes automatically snapshots your project before write_file, patch, rm, mv, and other destructive operations via a shadow git repository at ~/.hermes/checkpoints/store/.

> /rollback
Recent checkpoints:
1. 5 min ago — write_file: Dockerfile
2. 12 min ago — patch: app/main.py
3. 30 min ago — rm: temp.sqlite
> /rollback 1
Restored to checkpoint #1 (Dockerfile reverted)

/snapshot [create | restore <id> | prune] (alias /snap)

Manage global snapshots of Hermes’ own config and state. Unlike /rollback (file-level), /snapshot captures Hermes’ internal configuration.

> /snapshot create
Created config snapshot #4
> /snapshot restore 2
Config restored to snapshot #2

/stop

Immediately interrupt a running agent (background process, code execution, or LLM call).

Scenario 6: Background Tasks & Multi-Agent — Chat While Working

Hermes supports running inference in the background without interrupting your current conversation.

/background <prompt> (alias /bg)

Execute a prompt in the background. Output arrives as a system message without interrupting your flow.

> /background Analyze yesterday's Nginx logs, find anomaly IPs
Background task started (#42)
(You keep chatting...)
(Minutes later: Task #42 complete: Found 3 anomalous IPs)

/queue <prompt>

Queue a prompt for the next turn.

> /queue Summarize this page: https://example.com
Queued. Will run on the next turn.

/agents

Manage running agent processes — check status, stop agents.

> /agents
Background Agents:
  #42 bg-log-analysis — running (3 min)
  #48 queue-web-summary — queued

/moa <prompt>

Mixture of Agents mode — routes one prompt through the default model to multiple collaborating agents.

/subgoal <prompt>

Submit a sub-goal — directs the agent to focus on a specific sub-task.

/steer <instruction> / /goal <instruction> / /journey

  • /steer: Give the agent a mid-course correction without changing the overall goal.
  • /goal: Set or reset the top-level goal.
  • /journey: View the progress trajectory of the current goal.

Scenario 7: Skill Learning & Management — Make the Agent Remember Your Habits

/learn [what]

Create a reusable skill from anything — directories, URLs, chat content, or notes.

> /learn My Python project structure conventions
Creating skill "python-project-layout"...
Skill created. You can invoke it via /python-project-layout.

/skills [search | install | list | info <name>]

Discover, install, and inspect skills.

> /skills search web search
5 matching skills found:
  web-search-free    — Free web search (no API key)
  blogwatcher       — Monitor blog content changes
  research-workbench — End-to-end research assistant

/bundles

View skill bundles — groups of skills loaded with a single slash command.

/curator [status | run | pin | archive]

Manage the background curator’s automatic skill maintenance.

/suggestions

Review, accept, or dismiss curator-recommended automations.

/blueprint

Manage skill blueprints (pre-built skill templates).

Scenario 8: Configuration & Personalization — Make Hermes Yours

/config <key> [value] (CLI only)

View and modify configuration. Supports YAML dot-path notation.

> /config web.backend
Current value: ddgs
> /config web.backend searxng
Updated web.backend = searxng

/profile / /sethome

  • /profile: Show the active profile name and home directory
  • /sethome <path> (Gateway only): Reset Hermes home directory

/personality [name]

Switch personality packs. Hermes supports multiple preset personalities (professional, humorous, concise, etc.).

/skin [name] (CLI only)

Switch the CLI theme skin.

/indicator [on|off] / /statusbar [on|off] / /timestamps [on|off] / /verbose [on|off] / /footer [on|off] (CLI)

Toggle various CLI display elements.

/voice [on|off]

Toggle voice interaction mode.

/pet [list | <slug>]

Adopt a petdex mascot.

> /pet list
Available pets: cat, fox, dragon, owl, robot
> /pet cat
🐱 Cat activated! It will occasionally accompany you in chat.

Scenario 9: Tools & Plugin Management — Equip Your Agent

/tools (CLI only)

Interactive tool configuration wizard.

/toolsets (CLI only)

Fine-grained control over which tool sets load in each session.

/plugins [list | enable <name> | disable <name>]

Manage the Hermes plugin system.

/memory [search <query> | show <id>]

Search and inspect Hermes’ persistent memory.

/browser [list | close | inspect]

Manage browser sessions — view open pages, close tabs, inspect elements.

/kanban

Multi-profile collaboration board — manage tasks, links, comments.

Scenario 10: Platforms & Multi-Channel — One Hermes, Everywhere

/handoff <platform> (CLI only)

Hand the current session to another platform. Chat in CLI, continue in Telegram.

/platform [<platform>] / /platforms

View or switch the active gateway platform.

/topic [off | help | <session-id>] (Gateway only)

Control Telegram DM topic session mode.

/whoami

View your command access level on the current platform (admin / user / restricted).

Scenario 11: System Maintenance & Debugging — Daily Ops

/update

Update Hermes Agent to the latest version.

/version (alias /v)

Show the current version.

/debug

Generate and upload a debug report (system info + logs).

/reload / /reload-mcp / /reload-skills

Reload configuration, MCP servers, or skills without restarting Hermes.

/cron (CLI only)

Manage scheduled cron jobs.

/quit [--delete]

Exit the CLI. Use --delete to also remove the current session history.

Scenario 12: CLI Subcommands — Complete Toolchain

Beyond slash commands, 38 CLI subcommands cover installation, configuration, gateway management, and debugging:

Command Purpose Command Purpose
hermes setup First-time setup hermes install Install/update
hermes config Configuration hermes tools Tool config wizard
hermes gateway Gateway management hermes skills Skill management
hermes plugins Plugin management hermes cron Cron scheduling
hermes memory Memory management hermes model Model switching
hermes doctor Environment diagnosis hermes status Runtime status
hermes logs View logs hermes debug Debug reports
hermes dashboard Web dashboard hermes gui Desktop GUI
hermes login Nous Portal login hermes auth Auth management
hermes backup Backup & restore hermes security Security checks
hermes pairing Device pairing hermes webhook Webhook management
hermes profile Profile management hermes update Version update

Quick Reference by Scenario

Scenario Core Commands Priority
Session Management /new, /retry, /undo, /title, /clear ⭐⭐⭐⭐⭐
Search & Resume /sessions, /resume, /restart ⭐⭐⭐⭐⭐
Model Switching /model, /fast, /yolo, /reasoning ⭐⭐⭐⭐⭐
Context Compression /compress, /status, /usage ⭐⭐⭐⭐
Safety Rollback /rollback, /snapshot, /stop ⭐⭐⭐⭐
Background Tasks /background, /queue, /agents, /moa ⭐⭐⭐⭐
Skill Management /learn, /skills, /bundles, /curator ⭐⭐⭐⭐
Configuration /config, /profile, /voice, /pet ⭐⭐⭐
Tools & Plugins /tools, /plugins, /memory ⭐⭐⭐
Platform Switching /handoff, /platform, /topic ⭐⭐⭐
System Maintenance /update, /version, /debug, /reload ⭐⭐⭐
CLI Subcommands hermes setup, hermes config, … ⭐⭐⭐⭐⭐

Pro Tips

1. Chain commands

Combine /config + /model + /yolo to enter a fast-search mode:

> /config web.backend ddgs
> /model gpt-4o
> /yolo

2. Use aliases

Many commands have aliases:

  • /new = /reset
  • /compress = /compact
  • /snapshot = /snap
  • /background = /bg
  • /version = /v
  • /branch = /fork

3. Safety first

Keep /yolo off unless you’re sure. /rollback only works when checkpoints are enabled.

4. Master /compress focus

When a conversation drifts off-topic, don’t start a new session — use /compress focus <topic> to pull the focus back.

5. Keep skills tidy

Run /curator periodically to archive unused skills.

Conclusion

82 slash commands + 38 CLI subcommands sounds like a lot, but the core logic resolves to just 12 scenarios. For 80% of your daily needs, you only need 7 commands: /new, /retry, /undo, /model, /status, /compress, /sessions.

Master those seven, and you’ve already graduated from beginner to power user.

Remember: you don’t need to know all the commands. You just need the right ones for your workflow — and now you know where to find them.