---
title: >-
  How to Set Up an AI Agent That Runs a Clipping Channel (Hermes Desktop +
  Clipit, No Code)
summary: >-
  The literal no-code stack for an agent-run clipping channel: one open-source
  agent (Hermes Desktop) and one clipping engine (Clipit). You feed it a stream
  VOD, it returns scored and captioned clips, picks the best ones, writes
  titles, queues posts, and messages you for approval before anything goes live.
  No terminal required.
author: jordannneewbs
authorUrl: 'https://x.com/jordanneewbs'
category: Guides
difficulty: Beginner
readingTime: 8
date: '2026-06-23'
tags:
  - clipping
  - clipit
  - hermes-desktop
  - no-code
  - content
  - automation
  - discord
  - scheduling
  - skills
  - nous-portal
  - viral-score
integrations:
  - Hermes Agent
  - Clipit
  - Discord
  - Nous Portal
---

This is the literal setup I use. One open-source agent — [Hermes](https://x.com/NousResearch), free and MIT licensed — and one clipping engine, [Clipit](https://x.com/clipitdev). That's the whole stack.

The part that changed recently: **Hermes is a desktop app now.** Mac and Windows, normal installer, no terminal. Everything below happens inside the app.

**What you end up with:** an agent that takes a stream VOD, gets back scored and captioned clips, picks the best ones, writes titles, queues posts, and messages you for your approval before anything goes live.

## Step 1 — Install Hermes Desktop (2 minutes)

Go to `hermes-agent.nousresearch.com/desktop` and grab the installer — Mac (macOS 12+) or Windows (10/11). Run it like any other app. That's the whole step.

> Linux or terminal people: the CLI install is one command on the same site. Everything in this guide works there too.

## Step 2 — Sign in and pick a brain

First launch walks you through connecting a model. The easiest path is **Nous Portal** — there's a free tier, and the paid tiers include monthly credits, 300+ models, and built-in tool use (web search, browser, image gen) with zero extra setup.

If you already have a Claude/GPT subscription or an API key from any major provider, you can plug that in instead from settings.

## Step 3 — Have one real conversation

Before automating anything, verify the agent actually works. Ask it something it has to *do*, not just answer:

> "Look at my Downloads folder and tell me what's in it."

If it runs the task and holds a conversation, you're good.

Two things to notice while you're in there:

- **Persistent memory** — it remembers your projects and how it solved problems, across sessions.
- **Real capabilities** — it can browse the web, see images, and run tasks in the background. This is not a chatbot in a window.

## Step 4 — Put it in your pocket (Discord not required)

Hermes connects to **Telegram, Discord, Slack, WhatsApp, Signal, and email** — one agent, one memory, every surface. Set up the Discord connection from the app (it walks you through creating the bot and locks it so only you can message it).

This is the unlock for everything else: from here on, the agent is a contact you text. Approvals, status checks, "how did yesterday's clips do" — all from your phone.

## Step 5 — Connect Clipit (one click)

Now give the agent its clipping engine. Head to `clipit.dev`, click the profile circle (mine is yellow), open **Settings**, and go to the **API** tab. There's a section there literally called **"Connect Your Hermes Agent"** — hit **Quick Connect**.

That one click creates an API key with all the right permissions and hands you a prompt. Paste that prompt into Hermes and the agent does the rest itself — installs the Clipit skills, verifies the connection, done. You don't copy keys around or touch a config file; the agent sets up its own integration.

Two things worth knowing:

- There's a **Hermes Skills Pack** linked right on that page if you want to see what the agent just learned.
- API usage is charged to your clip credits balance — the agent spends the same credits you do, and you can see exactly what it's using from the dashboard.

## Step 6 — Teach it the clipping job

Here's the part that used to be technical and isn't anymore: you don't write config. You tell the agent the job in plain language, and Hermes turns the procedures it learns into **skills it keeps**. Mine breaks into three:

**The clip pipeline:**

> "When I give you a VOD link, run it through Clipit, wait for the scored clips, and export the top 3 to 5 by viral score."

Clipit's viral score is the whole reason this works — it gives the agent a number to make publishing decisions with, instead of needing a human to watch the footage.

**The niche scan:**

> "Every Monday, research which creators and clip formats are growing in my niche and where clipping bounties are posted. Send me a short memo."

**The posting job:**

> "For each approved clip, write the title and description in the channel's voice and schedule it."

Run each one manually a few times. The agent refines its own skills as it works — the pipeline after three weeks is genuinely better than week one. It learns your taste.

## Step 7 — Put it on a schedule

Scheduling is natural language too:

> "Every weekday at 9am, check for new VODs and run the clip pipeline. Post approved clips at noon and 5pm."

The agent runs unattended through the gateway and reports to Discord or the Desktop.

**One rule:** don't schedule anything until one clean manual run works end to end.

## That's the stack

One app, one engine, one phone. The agent owns the timeline work. You own the taste.

Building this in public with [Clipit](https://x.com/clipitdev) — the video production layer for AI agents. Featuring [Hermes by Nous Research](https://x.com/NousResearch).
