This is the first post on the Franklin blog. Here, you’ll find updates, behind-the-scenes notes, and thoughts about building privacy-first AI.
Stay tuned for more posts and news!
It’s wild how much can happen in just two weeks. When I look back at where Franklin was not long ago — fragile, half-wired, and constantly crashing — compared to today, it feels like I’ve been living inside a builder’s rollercoaster.
Two weeks ago, Franklin was barely alive. I’d fire up the container, only to watch it stumble and die with cryptic errors. Ports were mismatched, API keys missing, local tools broken. Half the time I didn’t know if I was debugging Python, Docker, or my own sanity.
It wasn’t glamorous. It was me, hunched over the terminal at midnight, scrolling through endless logs, muttering “why the hell isn’t this working?”
But then, the small wins started stacking up.
One night, after chasing down a port bug that had me swearing out loud, Franklin finally served its first clean response. Another day, the Web UI stopped throwing white screens and actually displayed a streaming reply. I’ll never forget the moment when I asked Franklin:
What’s in /memory?
And instead of crashing, it routed the question, executed the command, and calmly listed the files back to me. That was the first time it felt less like a pile of code — and more like an assistant.
If I had to name one thing that kept me sane, it was the test suite. I turned every bug into a test, every fix into a checkpoint.
Now I run the suite, and Franklin tells me the truth: green or red. It’s like having a safety net under the high-wire act. Without it, I’d still be flailing in the dark.
Building Franklin hasn’t just been about code. It’s been an emotional ride:
Some nights I went to bed frustrated. Others, I stayed up too late because I couldn’t stop tinkering. That’s the builder’s life — and I’m starting to love it.
Franklin isn’t finished. It still needs polish: smoother error handling, a proper settings menu, friendlier docs. But now, for the first time, I can see it. The vision is no longer abstract. Franklin is standing on its own two feet.
The next step is teaching it to run — and then, to share it with the world.
I didn’t start Franklin to build “just another AI app.” I started it because I wanted something different: local-first, private, mine. Watching it come alive these past two weeks has been proof that this crazy idea isn’t just possible — it’s happening.
And that’s a pretty amazing feeling.