We use local language models to assess and automate business solutions.
In development: A local AI diagnostic system that reads your machines' logs, explains problems in plain language, and proposes fixes for a technician to approve.
In development: Automated Document workflows.
Many AI tools work by sending your data to third-party servers. For system logs, that can mean machine names, user accounts and network details leaving your building.
We build on open-source language models that run on hardware you control. In our current prototype, the logs, the model and the alerts all stay on the local network.
Instead of digging through thousands of log lines, you get a plain-language explanation of what went wrong and a suggested fix. Nothing changes on your systems until a person approves it.
A central Linux host gathers logs from your Windows and Linux machines. A local language model reads them, flags hardware and software faults, and sends you a plain-language alert. Currently a working prototype in our own lab.
Learn More →We assess your repetitive IT and admin tasks and work out which ones a local model can handle reliably. Anything that changes a system is proposed first, tested in practice mode, and run only with a person's approval.
Learn More →Our core focus. We test open-source language models on lean, local hardware, study where they go wrong in real workflows, and refine prompts and checks to make their answers more dependable.
Learn More →One Linux box on your network does the watching, the thinking and the alerting. Here is what it does today, step by step.
Lightweight Grafana Alloy agents ship Windows event logs and Linux system logs to a central Linux host, where Grafana Loki stores and indexes them.
A local language model (currently Google's Gemma, running on the host itself) reads the logs and reports hardware and software faults in plain language, sorted by machine.
Alerts arrive on your phone or desktop through a self-hosted ntfy server. You can also ask a question, like "what happened on this PC in the last hour?", and get an answer drawn from the matching log lines.
The system proposes a repair script for a diagnosed problem, such as restarting a stuck print spooler. A practice mode shows exactly what would run without changing anything, and nothing runs for real until a person approves it.
The models run on local hardware, with no cloud AI service in the loop. That keeps sensitive logs in-house.
This is a working prototype running on our own lab machines, not yet a packaged commercial product. Today it analyzes system logs from Windows and Linux machines.
Your sensitive data should never be an afterthought. It is a core asset.
Relies on external APIs and third-party agreements.
Local infrastructure built inside your own network.
Black Pug Developments turns hands-on AI research into practical tools for keeping business systems healthy.
Christian Jerrett is a visionary with an extensive background in hardware and software who is currently developing and researching artificial intelligence solutions.
This website is itself a demonstration. Its first draft was generated by a local language model running on our own hardware, then refined by hand.
The same approach runs through everything we build: local models, real logs, and a person making the final call.
Our first conversation is a straightforward look at your current setup: what you want watched, where your data lives today, and whether a local system is a good fit.