Written for someone who has never opened a terminal. Do them one at a time. If a step fails, stop there and bring the error to the session rather than skipping ahead.
Before you start
You need: a Mac you are allowed to install software on, about 30 minutes, and a Telegram or Slack account. Nothing here needs a corporate IT ticket, and nothing here needs a credit card until the very last optional step.
GUIDE 01
Install Hermes Agent
This is the assistant itself. It is free and open source; you only pay for the AI model it uses, and only for what it uses.
Check one thing first
The macOS installer supports Apple Silicon Macs (M1 and later). Intel Macs are not a supported platform, so on an older machine use the terminal install path or skip this guide.
Pick your AI model. The wizard asks who you want to pay for model usage.
hermes model
Check it is healthy.
hermes doctor
Say hello.
hermes
The short path
If you would rather not pick a provider, one command signs you in and bundles the model plus the extra tools (web search, image generation, text to speech, browser):
hermes setup --portal
Optional extras
hermes gateway setup connects it to messaging platforms (the next two guides).
hermes dashboard opens a web admin panel with chat, settings and logs.
hermes desktop opens the desktop app if you installed via the terminal.
GUIDE 02
Connect it to Telegram
Telegram is the fastest way to talk to an assistant from your phone. It is a chat app, so there is nothing new to learn.
The 60-second path
Open the Hermes dashboard or desktop app and go to Messaging → Telegram.
Click Create with QR. Scan the code in Telegram, or open the link on the phone.
That is it. Hermes creates the bot, works out your Telegram user ID, saves both, and restarts the gateway.
The manual path, if you want to see the moving parts
In Telegram, search for @BotFather and open it.
Send /newbot.
Give it a display name (anything, for example "My Assistant").
Give it a username. It must be unique and must end in bot, for example lewis_assistant_bot.
BotFather replies with an API token that looks like 123456789:ABCdef.... Copy it.
Restart the messaging gateway so it picks up the change.
Optional polish
Message BotFather /setcommands to get a menu of slash commands in the chat.
Message BotFather /setprivacy if you want the bot to read every message in a group rather than only the ones that mention it.
Message BotFather /setuserpic to give it an avatar.
Treat the token like a password. Anyone holding it can control the bot. Never paste it into a chat, a slide or a screenshot. If it leaks, send /revoke to BotFather and set a new one. The ALLOWED_USERS line is what stops a stranger who finds your bot from using your assistant and your money.
GUIDE 03
Connect it to Slack
Slack is the work-appropriate surface: the assistant answers in a channel or a DM, and colleagues can be added to the same workspace later.
This uses Slack's Socket Mode, which means the bot dials out to Slack over a private connection. You do not need to expose anything on the internet, which is exactly why it is the right choice for a work laptop.
Step 1: let Hermes write the app definition
In Terminal, run:
hermes slack manifest --agent-view --write
It writes ~/.hermes/slack-manifest.json and prints instructions. That file already declares every permission, event and slash command the assistant needs, so you do not have to hunt for them.
Send the assistant a direct message in Slack. If it replies, you are done.
The two mistakes everyone makes. If it works in direct messages but stays silent in channels, the channel events are missing (Hermes' generated manifest includes them; a hand-built app usually does not). And if you ever change the app's permissions, you must reinstall it to the workspace, because Slack caches the old set.
GUIDE 04
Install Meetily
This is the meeting note-taker. It records and transcribes on your own machine, so a client's voice never leaves your laptop.
Install
Go to meetily.ai/downloads (or the GitHub releases page) and download the build for your machine. macOS and Windows have installers; Linux runs from a Docker or source build.
On a Mac: open the downloaded .dmg, drag Meetily into Applications, then open it from Applications.
On Windows: run the downloaded x64-setup.exe installer.
On first launch it asks for microphone and, on a Mac, system audio permission. Say yes to both. Without system audio it cannot hear the other side of the call.
First run
Download a local transcription model when prompted. Whisper and Parakeet models are offered; they are a few hundred megabytes and they stay on the machine.
Choose how summaries are generated. Ollama (local) keeps everything on the laptop. Alternatively point it at your own API key (Claude, Groq, OpenRouter, or any OpenAI-compatible endpoint), which sends the transcript out for summarising only.
Decide where the database lives, or accept the default.
Use it in a real meeting
Start the recording in Meetily before you join the call. It captures your system audio, so no bot joins the meeting and the client sees nothing unusual.
Let it transcribe live. You can keep working; the transcript builds itself.
When the call ends, click Generate Summary. Six summary templates are included out of the box.
Export. Markdown export is free; PDF and DOCX come with Pro. You can also copy the transcript straight out.
What is free and what is not
Community Edition, free: real-time transcription, audio and video import in 10 formats, retranscription, AI summaries, 6 templates, search and replace, Markdown export. Open source under the MIT licence.
Pro, US$10 per user per month (billed annually at US$120, 14-day trial, 30-day refund): speaker identification, custom AI model connector, custom summary templates, auto-detect meetings, PDF and DOCX export, and the MCP server, webhooks and CLI that let your assistant read your meetings.
Enterprise: from 10 licences upward, volume pricing and deployment help.
Where it fits with your assistant
On Pro, Meetily can hand a finished meeting to your assistant through MCP, a webhook when the recording ends, or the command line, which exports a meeting as Markdown, text or JSON. That is the difference between notes you have to find and notes that come to you.
Two honest caveats. Speaker labelling (who said what) is a Pro feature, so on the free tier the transcript is a wall of text. And analytics inside the app are opt-in by default, so leave them off if you would rather nothing at all leaves the machine.