Personal agent
Blacki
Blacki is a self-hosted personal assistant built on Google ADK, packaged to run on a small VPS without handing the whole experience to a cloud platform.
Goal
What I wanted to make
Make a personal agent that feels useful in everyday life while remaining affordable, portable, and fully under its owner’s control.
The problem
Why it needed to exist
Managed agent platforms add recurring cost, infrastructure lock-in, and limits around model choice. Small personal automations should not need a heavyweight cloud stack.
The solution
How the project answers it
Blacki combines Google ADK, Telegram, OpenRouter, Docker, CI/CD, and OpenTelemetry in a production-ready template that can run on a low-cost VPS. The Taylor Swift bridges experiment shows the point: ask naturally, then let the agent coordinate yt-dlp and ffmpeg into a finished artifact.
One wonderfully specific request
From a Telegram message to a 25-minute mix.
I gave Blacki the Taylor Swift tracks I love and asked for one continuous collection of their bridges. The agent had to turn that fuzzy request into a real media workflow.
What the agent worked out
- Find the bridge for each requested track.
- Download the source audio with yt-dlp.
- Cut the right moments precisely with ffmpeg.
- Stitch the excerpts into one continuous 25-minute track.
The moment it worked
i always wanted to do this. this is so exciting! embracing ai will change your life for sure.
— Chirag Patil (@VectorQl) July 9, 2026
ladies and gentlemen, it worked!!!!
— Chirag Patil (@VectorQl) July 9, 2026
super excited...