Over the last few months I've built a content factory of 13 AI agents - they publish posts, edit videos, chat in Telegram, handle clients. Along the way I learned to take apart other people's AI setups, fix them and optimize them. Now I help others
I work in two formats with full transparency about what's going on
Mode 1 - breakdown in chat
We create a chat for three: you, me, and my consultant agent
How it works
- you drop a question, a screenshot, a piece of config, an error log into the chat - anything that needs sorting out
- the bot answers first: reads it, suggests a solution, explains what and why
- I steer, help and comment
- I step in on tricky architecture questions or if the bot went the wrong way
- the whole conversation is handed over to you - you can reread it or send it to your own agent
A fit when you need to understand how to set things up right, compare approaches, break down your architecture, figure out why something doesn't work. No hands on your system - just breakdown and explanations
Mode 2 - my agent goes into your system
When you need not "explain" but "do". My agent logs into your VPS server or your Mac and gets to work
What it looks like
- you make a backup - we don't start without it, hard rule
- you create a separate user with limited permissions (5 minutes, I send the instructions in advance)
- we connect my agent, recording of the whole session turns on
- you see every command in the live feed before it runs: here's what we're about to do, here's why, here's what will change. You write "ok" → we do it
- dangerous stuff (rm, config edits, restarting prod services) - separately, only through a fully agreed plan
- after the work: the session recording file goes to you + a report on what was done
Who it's for: you built agents and got stuck. Or everything works - but it's expensive, slow, confusing. Or you want someone experienced to look at your configuration with fresh eyes and tell you what can be improved
About security
- the bot in the group has no shell access, doesn't read configs, doesn't know passwords
- over SSH I work under a separate user with limited permissions, not root directly
- the SSH key is generated separately for you and isn't used anywhere else
- the whole session is recorded + you get the file, you can review it after the fact
- Claude AI, which I work with - connection to Anthropic over TLS, data isn't used to train models, stored for 30 days
- any command that could break something is shown before it runs. Nothing launches without your "ok"
- you can close access at any moment with one command
What it looks like in practice
AI assistants I've set up now run for five people on their own servers. A clinic owner runs 6 agents, and I run 8 Claude Code agents myself every day
If you already have a system that is half-built and keeps breaking, I start by looking things over without changing anything: what works, what is broken, where the keys are, what spends money. Only then do I suggest what to finish and what is easier to rebuild. Setting up an assistant takes from one day; finishing an existing system takes 1–2 weeks depending on its size
I work on your server and your AI subscription, so the bill for the models goes straight to the provider and your keys never leave. The clinic owner's voice agent, for example, thinks on his own ChatGPT subscription
An example from my own system: my personal agent Jes has been running for 134 days, does 10 scheduled jobs on her own and hands tasks to 7 other bots. Anything risky, paid or public waits for my yes. I build the same safety rules into other people's systems when I bring them to a working state
The chat breakdown suits you if you want to do it with your own hands and understand it. Work on your system suits you if there's no time and it just needs to work
Message me on Telegram → @matthewjester
In two sentences: what your system is and where you're stuck - we'll fix it and set everything up
The task-review format is described on the AI consulting page