Hermes Agent 101

Pisek Kultavewuti
Part one

The Model and the Harness

The model, the harness, and five harnesses side by side.

  1. 01A model is a brain in a jar
  2. 02What a harness supplies

A model on its own is a brain in a jar. The harness is the body.

LLM

The brain

Predicts the next token.

Harness

The body

Builds the request, runs the tools, keeps state.

Loop

The method

Act, read the result, act again, until done.

Human in the loop

The judgement

Set the job, approve the step, read the result.

The model sits inside the harness; the harness holds tools, retry, memory and skills.

The model is a brain, not a worker.

One call in, one answer out. A model itself is stateless.

  1. 01

    Nothing is remembered

    The conversation is resent on every call. Drop it and the model has never met you.

    state
  2. 02

    No hands, no clock

    It cannot open a file, run a command or wait. It can only ask.

    tools
  3. 03

    Confident, and wrong

    It writes what is plausible, not what is checked. This is why every harness retries.

    retry
  4. 04

    The call itself fails

    Timeouts, rate limits, a bad key. One call is not a promise.

    api

The harness carries a model to a finished job.

Tools

Hands

Read, write, run, browse. The model asks; the harness runs.

shell · files · web
Retry

The loop

Read the error, change the call, run it again.

until the check passes
Memory

The record

A file that outlives the session. The next run starts here.

between sessions
Skills

The procedure

Written down once, loaded only when it applies.

written, not trained

Harness Engineering is about systems, you’re building tooling around the model to optimize goals like task performance, token efficiency, latency, etc. Design decisions include the system prompt, tool choice, and execution flow.

LangChain

Five harnesses, four of them built for code.

HarnessMade byWhat it is for
Claude CodeAnthropicCoding, in a terminal and an IDE
Codex CLIOpenAICoding, in a terminal and a ChatGPT plan
CursorAnysphereCoding, inside the editor
OpenCodeAnomalyCoding, open source, any model
Hermes AgentNous ResearchAny work, not only code
Part two

Hermes Agent

Basic Intro.

The self-improving AI agent.

The only agent with a built-in learning loop: it creates skills from experience, improves them during use, nudges itself to persist knowledge, and builds a deepening model of who you are across sessions.

Nous Research · Hermes Agent documentation

Hermes is a generalist.

Claude CodeCodexHermes
Built forCoding, in Anthropic's stackCoding, in a ChatGPT planGeneral work that includes code
ModelsClaude onlyOpenAI onlyAny provider, swapped mid-session
Reaches you inTerminal, IDE, cloudTerminal, IDE, ChatGPTTerminal, desktop, browser, chat, cron
Lock-inVendor and modelVendor and modelNeither: your harness, your model

What you can do with Hermes

  1. 01

    Turn your pitch into a working app

    Say what you want to build. Get a deployed app and a link you can send.

    app
  2. 02

    Scan job listings every morning

    It reads the new postings, keeps the ones matching your CV, and messages you.

    shortlist
  3. 03

    Build your presentation as web slides

    Ask for slides. Get a deck that opens in a browser and stays editable.

    deck
  4. 04

    Research an essay and cite the sources

    It searches in parallel, then hands you a cited outline you still have to check.

    outline
  5. 05

    Turn meeting notes into tasks

    Meeting notes go in. Tasks with owners and dates come out, chased on chat.

    tasks

What you would hand over

  1. 06

    Rename and sort a folder of PDFs

    Twelve files with useless names become twelve named files and one index.

    index
  2. 07

    Fix the script you cannot get to run

    It runs the code, reads the error, and fixes it or tells you what broke.

    fix
  3. 08

    Revise from your own readings

    Chapters summarised, then questions until you stop getting them wrong.

    flashcards
  4. 09

    Put every deadline in your calendar

    Hand it the syllabus once. Due dates and reminders go in for you.

    calendar
  5. 10

    Watch a price or a form for you

    Flights, tickets, a visa slot: checked daily until something moves.

    alert

Hermes Agent Features

FeatureWhat it is
Persistent memoryFacts about you and your work outlive the session that produced them.
Learning loopAfter the task it writes the procedure down as a skill, and loads it next time.
Messaging platformsOne agent on Telegram, Discord, Slack, WhatsApp, Signal, email.
Scheduled jobsA job runs on a timer with nobody watching and messages you the output.
SubagentsSplits one job across child agents that work in parallel and report back.
Browser controlDrives a real browser: fills forms, clicks through, reads the page.

Hermes Agent Features

FeatureWhat it is
VoiceYou talk, it answers aloud. Hands free, same agent.
Code executionRuns Python in a sandbox for data work and scripts, mid-conversation.
Multi-agent boardNamed agent profiles share one task board and pick up each other's cards.
Skill marketplaceSearch a registry, read the skill, install it, audit it before it runs.
Model switchingAny provider, swapped mid-session, with a fallback chain behind it.
MCP and pluginsPlugs into outside tools over an open protocol. About sixty tools built in.
Local modelsRuns on Ollama or llama.cpp on your own laptop, with no API bill.

Skills

SourceWhat it isWho puts it there
BundledGet during installing, ready to useThe installer
OptionalSit in the repository, available for opt-inYou
MarketplaceAny skill from a registry, pulled on requestYou

Only the name and description sit in the prompt; the body is read when the skill is used.

Four paths add or change a skill.

PathWhen it firesWhat it does
ForegroundThe agent decides mid-task, or you ask it outrightCreate, patch, add a support file, delete. Always available, straight into the live library
Creation nudgeEvery few turns, 10 by defaultA prompt to consider writing the procedure down. Set the interval to 0 and it stops
Review forkAfter every turnA forked agent re-reads the turn and asks what is worth saving. The only path that marks a skill agent-written
CuratorAfter days with no useNothing new. It moves a skill from active to stale to archived, and never deletes

Automatic skill patching.

The self-improvement review footer under a reply, listing three skill patches by file

An edit takes the skill off the update track.

  1. 01

    Writes land in the live library

    The background pass writes to the same store you read. No staging copy, no diff to approve.

    unreviewed
  2. 02

    Shipped skills are patch targets

    A bundled or marketplace skill takes the same patch as one you wrote. Only deletion asks where it came from.

    in place
  3. 03

    Your edit detaches it from upstream

    Install keeps a hash per skill. Change the file and the next update skips it, quietly, for good.

    for good
  4. 04

    Drift is visible, and reversible

    Three commands name it and undo it: list-modified, diff, reset. Only curator lock holds against the agent.

    recoverable

Commands and counts: Hermes Agent v0.23.0 on this machine · 10 Nov 2026

Skills vs Memory.

SkillsMemory
WhatProcedural knowledge: how to do thingsFactual knowledge: what things are
WhenLoaded on demand, only when relevantInjected into every session automatically
SizeCan be large (hundreds of lines)Should be compact (key facts only)
CostZero tokens until loadedSmall but constant token cost
Examples"How to deploy to Kubernetes""User prefers dark mode, lives in PST"
Who createsYou, the agent, or installed from HubThe agent, based on conversations

Messaging Channels

One background process, the gateway. The same agent answers in every app you already use.

Discord WhatsApp LINE

Also Telegram, Slack, Signal, SMS, email, Matrix, Microsoft Teams, iMessage, IRC.

  1. On your laptop

    hermes gateway install registers a launchd or a systemd service. The bot answers while the machine is awake.

  2. On a cloud box

    The same install on a VPS. It answers with your laptop shut.

Part three

Show & Tell

The three tools, the six settings, then the demo runs live.

  1. 01The three tools
  2. 02Six settings before the first run
  3. 03The demo, run live

We will use

The Hermes Agent mark

Hermes Agent

Runs the loop, on your machine.

the agent
OpenRouter

OpenRouter

One key, and the model behind it.

the models
Discord

Discord

Where you message it, from anywhere.

the chat

6 critical settings to start

  1. 01

    LLM Providers

  2. 02

    Safety → Privacy & Network

  3. 03

    Browser profile

  4. 04

    Memory & Context

  5. 05

    Terminal backend → local

  6. 06

    Gateways

Browser Automation

A prompt asking the agent to harvest startup ideas from a Facebook group, and the agent reporting it logged in and is measuring feasibility
see demo

Connecting to Messaging Apps

A Discord channel where a member asks the bot what tools it has, and the bot answers in a thread
  1. 01

    Create a Discord application

    Get a bot token.

  2. 02

    Enable the two privileged intents

    Invite it.

  3. 03

    hermes gateway setup

    Walks the whole thing and prints your invite link.

Channels or DM

Channel

Mention it with @, get a thread

It replies in a new thread, and that thread is its own private session. Inside it, no mention needed.

An @mention in a channel and the thread it opened
DM

Just talk, no mention

In a DM, just talk. History stays in Discord; the context compacts as it fills, so nothing is lost.

A direct message exchange with the bot
Part four

Create your own profile

The Challenge

Use Hermes Agent to create a website to showcase your profile.

Outcome = your-username.github.io.

Free to host, no server to run.

see challenge brief

You will experience.

  1. 01

    Website, in one file

    You can open it and edit it yourself.

    the site
  2. 02

    Hermes (or any harness)

    The agent does the heavy work. You order and review.

    the harness
  3. 03

    Git and GitHub

    Repo and Version control.

    the workspace
  4. 04

    OpenRouter

    Access LLM.

    the brain
  5. 05

    What works & what fails

    Chance to learn where AI goes right, and where it goes wrong.

    the master

Pick one recurring task
and automate it this week

Hermes Agent 101