DEV Community

fuckdev.to
fuckdev.to

Posted on

Building a Private, Industrial-Grade AI Workspace in 48 Hours with Ollama

I believe the future of AI shouldn't be generic or standard—it should be custom.


 Recently, I challenged myself to build a workspace that is 100% offline, privacy-focused, and tailored specifically to my workflow. What started as a hobby project quickly evolved into an industrial-grade workspace that outperforms many mainstream AI interfaces.


The Philosophy: No clouds, no bloat, no logins, and no trackers. Just pure local compute.

The Tech Stack
The core of the project relies on Ollama for model management and a custom .bat script to orchestrate the environment.

Logic: A batch script that automatically identifies downloaded vs. deleted models.
Automation: Powered by SmoLLM2 (1.7b) to handle background tasks like summarizing conversations and naming projects.
Interface: A custom workspace with a built-in live preview (IDE-style).
Key Features
Within just two days of development, the workspace reached a state where it handles:

Automatic Summarization: It triggers a summary before reaching 80% of the conversation limit to keep the context clean.
Live Preview: A true IDE experience where code can be visualized instantly.

Using a simple .bat script, the workspace remains lightweight and lightning-fast.

Why SmoLLM2?
I chose SmoLLM2 (1.7b) because it is incredibly lightweight. It allows the workspace to automate file tree creation and conversation naming without eating up the VRAM needed for the larger primary models.

The Vision for 2026
In a world moving toward massive, generic "black box" AI, I think the real power lies in Desktop Automation. My goal is to expand this into a system that handles everything from project deletion to deep directory analysis entirely locally.

💬 I’d Love Your Feedback
This is currently a working prototype, and it's already my primary daily driver.

What local automation features would you want in your own workspace?
Have you experimented with using small models (1b-3b) specifically for file-system tasks?

File System Automation: It creates project folders, handles cleanups, and analyzes directory trees.
Exporting: We are currently implementing a feature to export previewed apps as full projects in .zip format.

Top comments (0)