
Five lessons from using Glitch to find businesses and set up Instantly
What Pixelhop learned when Glitch missed a known-good business and configured an Instantly follow-up incorrectly.
Gemma
Your own AI teammate, set up and looked after.
We set up Hermes Agent on a small computer that runs in your office, connect it to Slack, Teams, WhatsApp or Telegram, and give it useful jobs from day one. It's the same setup we use to run our own agent, Glitch, every day.
Talk to usInstalling Hermes is the easy part. What makes an agent useful is the computer it runs on, the chat apps it lives in, what it knows about your business and the jobs it does without being asked. We set up all of it.
We set up a small, quiet Linux mini PC with 32 GB of memory, test it and post it to you. It plugs into your network, so the agent, its memory and its files live in your office rather than on a shared platform. We look after it remotely over a secure connection.
Installed, connected to your Codex subscription and running as a service that restarts itself if anything falls over. Updates are tested before they go on.
Slack, Microsoft Teams, WhatsApp, Telegram or Discord. Your team talks to the agent where they already work, and it reports back there too.
We spend a session with you on your business, customers, services and the way you write, and put it into the agent's memory. It starts out knowing the basics and keeps learning as you use it.
Jobs it does on a schedule without being asked: researching new leads each morning, a twice-weekly sweep of who you owe a reply, and a weekly check-in on what's moving. We tune them on your real data.
You decide what needs your approval. By default it drafts emails and messages for you to approve, and it doesn't send anything on its own. We finish with a handover session so your team knows how to work with it.
Each morning Glitch looks up every new sign-up to our product Chat Thing, skips the joke names and repeat sign-ups, and writes a personal email draft for the rest, with the sources it used. We approve or skip each one with an emoji in Slack.
When someone uses one of our free tools, Glitch works out who they are, rates the fit and finds a public way to contact them. It checks our inbox and CRM first, which once stopped us cold-emailing a lead who already had a sales call booked.
On Monday mornings and Friday lunchtimes it reads our shared inboxes and sorts every open conversation into replies we owe, people worth chasing and threads to leave alone, each with an owner. It also says what it couldn't check.
It knows our services, prices, past conversations and how we write, so a quick question in Slack gets an answer that fits. It saves standing preferences too, like keeping questions short while we're writing.
When we launched a new Chat Thing feature, Glitch drafted the blog post, the launch email with subject lines to test and the social posts in one afternoon, ready for us to review.
Tell us about your use case. We'll come back with a straight answer about whether it's something we can help build.
It gets things wrong too. We wrote up the mistakes and what we changed in five lessons from using Glitch. Want an agent that does more than this? See AI agent development.
A few projects where we put this approach to work. Click through for the full story behind each one.

What Pixelhop learned when Glitch missed a known-good business and configured an Instantly follow-up incorrectly.
Gemma
How a free SEO checker started surfacing potential leads, and the human-reviewed Glitch workflow we built around it.
Gemma
Our first real-world test of Anthropic's Fable 5: running inside Claude Code, it autonomously rebuilt a messy, already-deployed prototype onto our production stack in a single afternoon. Hugely impressive and near-faultless, but slow and notionally pricey, which raises a real question once API pricing lands.
JozefA one-off setup fee that includes the computer, then a monthly care plan that keeps the agent running. Your ChatGPT plan with Codex is paid directly to OpenAI.
£2,500 + VATone-off
Everything needed to get a working agent into your team chat, doing useful jobs.
£350 + VATper month
Looking after the agent remotely, so it keeps working after the first week.
Add-ons are quoted at our day rate of £650 + VAT. If you want an agent built around your own product or systems, that becomes AI agent development.
Hermes Agent is a free, open-source AI agent from Nous Research. It runs on a computer, remembers what it learns, does jobs on a schedule and talks to you through chat apps such as Slack, Teams and WhatsApp. The software is free; you pay for the computer it runs on and the AI model behind it.
Founders and small teams in the UK who know what AI can do but don't have the time or technical help to set up an always-on agent themselves. It suits businesses that get regular leads or enquiries, have busy shared inboxes, or keep forgetting the same admin. If you want an AI assistant that does real work in Slack, Teams or WhatsApp rather than another chat window, it's for you.
The care plan is £350 + VAT a month. On top of that you pay OpenAI for a ChatGPT plan that includes Codex, which powers the agent. There's no hosting bill, because the agent runs on the computer we send you, and no other running costs unless you add paid tools. See What it costs for the detail.
No. The £2,500 + VAT setup fee includes a small Linux mini PC with 32 GB of memory and a 1 TB drive. We set it up, test it and post it to you, you plug it into your network and power, and we do the rest remotely. If the hardware ever fails, we restore your agent onto a replacement from its backup.
Yes. We recommend a ChatGPT plan with Codex, which is what runs Glitch. The subscription is in your name and paid directly to OpenAI, so you stay in control of it. We connect it to your agent during setup.
Slack, Microsoft Teams, WhatsApp, Telegram and Discord are the ones we set up as standard, and Hermes supports others including Signal and email. WhatsApp connects by linking a phone number, the way WhatsApp Web does, so we always set it up on a new number and never your personal one.
Your agent runs on its own computer on your network, and it can only reach the tools and accounts you connect. By default it drafts emails and messages for your approval, and it doesn't send anything on its own. We're UK GDPR compliant, and at setup we agree what data the agent can see and who can access it.
It will, now and again. Logins expire, updates change things and scheduled jobs can fail quietly. We've had all of these with Glitch, which is why the care plan includes monitoring and fixes. You can also message us whenever something looks wrong.
Yes. It's free, the documentation is good, and if you're comfortable in a terminal you can have it installed in an afternoon. The slower part is preparing and securing the hardware, connecting your chat apps and tools, giving it a memory of your business, writing routines that do real work and keeping all of it running. That's the part we do.
We ran an OpenClaw agent before we moved Glitch to Hermes, and Hermes is what we've used every day since August 2026. It keeps a memory between conversations, turns jobs it repeats into skills it can reuse and runs scheduled work by itself. Putting every client on the same Hermes setup means we know it inside out and can look after it properly.
Yes. If you later want custom tools, a connection to your own product or an agent your customers talk to, that becomes AI agent development, and the agent you already have is a good place to start from.
Yes. We're based in Newquay, Cornwall. We post the computer to you and do the rest remotely, with the business session and handover on video calls. Most of our clients are elsewhere in the UK.
Tell us what you'd want it to do first. We'll tell you straight whether an agent is the right fit.
Start a conversation