Tap the right side for the next slide · tap the left side to go back · swipe or arrow keys work too

FEATURES • SETUP • EXAMPLES

Hermes Agent

Recent Updates & the Complete Feature Guide

Dr. Farshid Pirahansiah • AI & Computer Vision Engineer
Every feature, with a runnable example — pirahansiah.com

Recent Updates (2026)

🚀 Core

• 21+ messaging platforms (19 native + IRC & Teams via plugins).

• Git worktrees for safe parallel agents on one repo.

• Checkpoints & rollback via shadow git snapshots.

• Import from Claude Code / Codex CLI in one command.

🧩 Extensibility

• ~90 bundled skills + ~60 optional installable.

• Profile distributions — share a whole agent.

• Managed Scope — admin-pinned immutable config.

• Egress proxy / iron-proxy for credential injection.

Install: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Setup & Configuration

# Install (uv + python venv + launcher)
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

# Interactive setup wizard + provider picker
hermes setup          # model | tts | terminal | gateway | tools
hermes model           # pick provider + model interactively
hermes doctor          # dependency + config health check
hermes status --all    # component status

# Config is set via CLI (never hand-edit the YAML)
hermes config set model.default Qwen3.5-4B-OptiQ-4bit

Providers & Model Aliases

☁️ Cloud

OpenRouter, Anthropic, OpenAI, Gemini, DeepSeek, xAI, Qwen, Nous.

hermes auth add gemini

🖥️ Local

oMLX, llama.cpp, Ollama, vLLM — any OpenAI-compatible endpoint.

model.base_url: http://127.0.0.1:8000/v1

🔀 Fallback

Auto-rotating credential pools + fallback chain.

hermes fallback add openrouter

# User model aliases (resolved before built-ins)
hermes config set model.aliases.cheap gemini/gemini-2.0-flash-lite
# In chat: /model cheap   (session) or --global (persist)

Skills — Procedural Memory

📚 Catalog

• ~90 bundled skills: arxiv, jupyter-live-kernel, docx, pdf, architecture-diagram, github, google-workspace.

• ~60 optional installable from the hub.

hermes skills browse / search QUERY

⚡ Create Your Own

Save a reusable procedure as a skill; it auto-loads when the task matches.

skill_manage create cv-yolo \
  --cat cv
# writes SKILL.md → loads on match

New Skills Installed & Validated 2026-09-06

Installed via hermes skills install / URL, scanned, and verified (valid SKILL.md frontmatter, enabled).

Web & Research

• agent-reach — internet eyes, no paid API
• youtube-full — transcripts/search
• defuddle — clean pages → Markdown
• resemble-detect — spot AI fake media

Build & Engineering

• using-agent-skills (addyosmani)
• setup-matt-pocock-skills
• make-interfaces-feel-better
• humanizer — strip AI tells

Agents & Ops

• browser-harness — drive real browser
• i-have-adhd — action-first replies
• loopy + loop-library — repeatable loops
• skillclaw — self-improving skills

# Install example (hub id or raw SKILL.md URL) hermes skills install skills-sh/panniantong/agent-reach/agent-reach -y hermes skills install https://raw.githubusercontent.com/resemble-ai/detect-skill/master/SKILL.md -y # 14 skills installed & validated · 15 community/url total now active

Blocked by security scan (dangerous verdict, --force cannot override): codebase-memory-mcp, openmontage, composio, claude-mem, rlaope/oh-my-hermes. Plugin-only (not flat SKILL.md): witt3rd/oh-my-hermes, agent37/minions. Use curated repo paths or audit before install.

Persistent Memory

🧠 What It Remembers

• Who you are, your role, environment facts.

• Stable conventions (no task home).

• Lab GPU inventory, SSH nodes, research domain.

hermes memory setup|status|reset

🔁 Across Sessions

Injected into every turn — small, high-signal, declarative facts only.

Procedures → skills; facts → memory.

Hard cap enforced (character budget) — auto-consolidates.

Multi-Agent Orchestration

# Spawn isolated subagents in parallel (each = own context + tools)
delegate_task(tasks=[
  {"goal": "Fine-tune YOLOv11 on /data/coco-subset, export ONNX",
   "context": "epochs=50, device=cuda:0"},
  {"goal": "Run SAM2 over /data/video/*.mp4, emit mask JSON"},
  {"goal": "Profile GPU memory under 50 streams with py-spy"}
])
# Coordinator merges verified summaries — no manual glue.

Also: hermes moa (Mixture-of-Agents), hermes kanban (work-queue), --worktree (isolated checkouts).

Bot Mode — Named Agents

🤖 Roster of Bots

Each bot = own chat, role, model, memory, skills, avatar.

Bots run routines, share group chats, message each other.

⚙️ Create One

hermes bot create cv-alerts \
  --platform telegram \
  --group "CV War Room"
# bot runs autonomously, posts results

Messaging & Group Bots (21+ Platforms)

Channels, groups, bots.

Discord

Slash commands.

Slack

Webhook alerts.

Baileys + Cloud.

iMessage

Photon adapter.

Matrix

+ IRC, Teams, more.

hermes gateway run          # launch messaging gateway
hermes auth add telegram     # OAuth / token once
hermes send "Eval done: mAP 0.94" --platform telegram

Cron & Scheduling

# Nightly CV eval at 02:00, deliver to Telegram
hermes cron create "0 2 * * *" \
  --name "Nightly CV Eval" \
  --script nightly_eval.py \
  --deliver telegram

# Human-readable schedules also work
hermes cron create "every 2h" --name "Health Ping" --no-agent --script ping.sh
hermes cron list | pause | resume | run ID

Survives reboots; runs headless; notifies on completion or failure.

Research Assistants

ArXiv

Search + PDF parse → BibTeX.

Citations

grounded-citations: zero hallucination.

Wiki

llm-wiki knowledge base.

Jupyter

Live kernel for experiments.

LaTeX

docx / pdf skill for papers.

Diagrams

architecture-diagram SVG.

Lit sweep → code reproduce → experiment → draft, all agent-driven.

Profiling & Debugging

⚡ Performance

• nvidia-smi dmon loops captured by agent.

• py-spy dump for Python GIL stalls.

• torch.profiler → flamegraph SVG.

• throughput sweeps (1→32 batch).

🐛 Systematic Debug

• systematic-debugging skill: 4-phase root cause.

• checkpoints: shadow-git snapshots, /rollback.

• hermes logs -f errors for live traces.

Artifact Management

📦 Produced

• Weights (ONNX, GGUF, safetensors).

• Benchmarks: CSV + interactive HTML.

• Diagrams, datasets, diff reports.

☁️ Stored & Traced

• Auto-push to HuggingFace Hub.

• Git-LFS for large binaries.

• Every artifact tagged: commit + eval hash.

• Reproducible via skill_manage replay.

Surfaces & Power Features

Desktop

Native GUI (macOS/Win/Linux).

Dashboard

Web admin + chat.

TUI

Ink terminal UI.

Proxy

OpenAI-compatible local proxy.

MCP

Native + catalog servers.

Profiles

Isolated configs/skills.

WebhooksACP / IDEPetsSkinsTTS/STTComputer UseImport Claude/Codex

Complete Feature Catalog

Autonomous tool-callingPersistent memorySkill system (~90+~60)Multi-agent (delegate)MoAKanban queueGit worktreesCheckpoints/rollback35+ providersLocal LLM (oMLX/llama.cpp)Model aliasesFallback poolsBot mode21+ messaging platformsGroup botsCron schedulingWebhooksMCP serversResearch: ArXiv/BibTeXGrounded citationsJupyter live kernelProfiling (GPU/py-spy)Systematic debuggingArtifact lineageHF pushDesktop GUIWeb dashboardInk TUILocal proxyProfiles & distributionsSecrets (1PW/Bitwarden)Managed scopeSkins & petsVoice TTS/STTComputer useACP / IDEImport Claude/CodexEgress proxy

→ Every feature shown above has a working example in this deck.

One Agent. Every Workflow.

✔ Setup once — config + skills + memory persist forever.

✔ Orchestrate — multi-agent swarms, bots, cron loops.

✔ Operate anywhere — CLI, GUI, web, 21+ messengers.

✔ Ship traced artifacts — reproducible, push-button.

pirahansiah.com/farshid/content/research-tools.md

Validated Skills Library installed & scanned

All installed via hermes skills install, security-scanned, and verified (valid SKILL.md, enabled).

Web & Research

agent-reach · youtube-full · defuddle · resemble-detect

Build & Engineering

using-agent-skills · setup-matt-pocock-skills · make-interfaces-feel-better · humanizer

Agents & Ops

browser-harness · i-have-adhd · loopy · loop-library · skillclaw

# list active skills hermes skills list # 14 validated community/url skills now active

Command Cheat Sheet DESK · CLI · MSG · SHELL

Everyday commands — Desktop, CLI chat, messaging, shell. (Verified vs current Hermes source, Sep 2026.)

Sessions & Context

/new · /resume · /sessions · /title · /branch · /compress · /context · /status

Control Work

/queue · /steer · /bg · /btw · /agents · /stop

Goals, Loops & Plans

/goal · /subgoal · /heartbeat · /loop · /plan · /review · /refine

Models & Behavior

/model · /moa · /personality · /reasoning · /fast · /approvals · /yolo · /busy · /voice

Skills, Memory & Tools

/skills · /learn · /memory · /init · /tools · /browser

Automation & Coord

/cron · /suggestions · /blueprint · /kanban

Inspect, Recover & Fix

/retry · /undo · /save · /diff · /rollback · /usage · /debug · /help

Messaging / Gateway

/sethome · /topic · /commands · /approve · /deny · /pause · /platform · /restart

Terminal Essentials

hermes / hermes chat · -z "prompt" · model · status · doctor · gateway status · --safe-mode · update

# repeat-work modes /goal = work until objective met /loop = repeat w/ stop conditions /heartbeat = one recurring check /cron = durable schedule outside chat # type / + letters for autocomplete; /help for full list

Thank You

Hermes Agent — Updates & Complete Feature Guide

All Presentations • Notes & Guides • GitHub • Official Docs