Hey there, fellow terminal gremlins 👋
I'm Jay—aka Ghotet, chaos-friendly solo dev, digital necromancer, and long-time tinkerer of systems no one asked for. Today I’m sharing how I stitched together a fully offline, self-hosted ChatGPT clone using open source parts, some bash glue, and a healthy disrespect for cloud dependency.
TL;DR: I wanted an AI assistant that works without internet, plays nice with local tools like Stable Diffusion and SearXNG, and lives on my own domain. Also, my cats keep breaking it.
🧠 The Stack: Built from Curiosity and a Bit of Spite
🔍 Core LLM — LM Studio (running Mythalion 13B)
I’m using LM Studio as the language engine, running locally via GGUF models. Right now it’s dialed to Mythalion 13B for its mix of smarts and creative leanings. The whole thing runs headless from a custom launcher script. No API keys, no cloud, no telemetry.
🎨 Visuals — A1111 (Stable Diffusion)
For visual generation and character concepting (yes, including bikini armor if I want it), I wired in Automatic1111, running fully offline. API access is enabled so prompts can be piped in directly from the chat interface.
🌐 Web Search — SearXNG (Self-hosted)
To enable research and general brain-extension, I slotted in SearXNG. It runs locally, routes search queries anonymously, and feeds results into the stack when needed. Originally tried Docker, but now it's running through my venv setup for tighter control and portability.
🕸️ Web Frontend — Terminal-style Interface @ ghotet.com
Everything is wrapped in a custom retro terminal UI hosted at ghotet.com. Think early-2000s cyberpunk console vibes. Currently private—no guest login just yet—but it's functional and fast.
💡 Why I Did This
Because I like my tools local, moddable, and entirely mine. I wanted:
A personal chatbot that actually lives on my machine
Visual generation without round-tripping to some cloud GPU farm
Web search that doesn’t log me into a panopticon
A unified front-end I can bend to my will
Also, let’s be real: I’m a sucker for a terminal that feels alive.
⚙️ What’s Working
✅ LM Studio launches cleanly and routes input/output
✅ A1111 runs with API access and feeds visuals into the flow
✅ SearXNG is now portable and integrated into the chat stack
✅ The full interface is browser-accessible via my domain
✅ Modularity is in place: every part can be swapped or upgraded
📱 What’s Next
🔧 Remote Access
I’m working on a remote control bridge, so when I forget to shut down the stack before leaving, I can log in from my phone and reset it. This has become necessary because my cats have discovered keyboard inputs and enjoy triggering hotkeys that collapse my entire setup.
🔊 TTS (Maybe)
I’ve been testing offline text-to-speech tools, but most open options either lag too much or sound like a haunted Speak & Spell. If I find one fast enough for real-time chat, I’ll wire it in.
🧩 Bonus Project — ARG Terminal @ ghotet.dev
I’m also migrating my alternate reality hacker terminal to ghotet.dev. It’s part portfolio, part digital rabbit hole. Expect binary rain, cryptic prompts, and a few unsettling vibes. Less functional, more for fun. I haven't fully set-up the new domain yet so it is likely not up at the time of writing.
🗣️ What About You?
I’d love to hear from other tinkerers—if you’ve built your own local AI stack, hacked together a weird interface, or just like the idea of owning your tools, drop a comment.
Here’s a few sparks to get the thread moving:
What’s your dream self-hosted AI setup look like?
Got any cool tools or models I should try out?
Is there a non-creepy TTS stack that doesn't sound like it escaped from a dial-up modem? Asking for a friend.
Am I the only one whose pets keep triggering system-wide mayhem, or is that just part of the dev life now?
Let’s chat. I’m around. And if your comment is interesting enough, maybe I’ll wire your idea into the next update 😏
If you have any questions about me, feel free to ask or follow me on dev.to or github.com @ghotet.
Top comments (0)