Show HN: I'm 15. I built an offline AI Terminal Agent that fixes errors

by taklaxbr • 7 months ago

4 10 7 months ago

taklaxbr5• 7 months ago

Hi HN

I'm a 15 year old developer from Turkey. I previously shared an earlier version, but today I'm releasing ZAI Shell v7.0 (Agent Edition), which is a massive overhaul.

I built this because I was frustrated with existing AI CLI tools that break easily or require complex setups.

What makes v7.0 different?

Self-Healing: If a command fails (e.g., encoding error or wrong shell syntax), it analyzes the stderr, switches strategies (CMD -> PowerShell -> WSL), and retries automatically.

Offline Capable: It can run fully offline using a local Microsoft Phi-2 model if you don't want to use Gemini API.

Hybrid Agent: It's not just text anymore. It can now see (Vision), search the web (DDGS), and even control the mouse/keyboard (GUI Automation) to complete tasks that terminal commands can't handle.

Modular Architecture: If you don't have PyTorch or PyAutoGUI installed, it doesn't crash; it just disables those features and works as a lightweight shell.

I wrote the core logic in Python using subprocess for shell management, socket for P2P sharing, and chromadb for memory.

The code is open source. I'd love to hear your feedback on the new architecture!

Repo: https://github.com/TaklaXBR/zai-shell

anandvc• 7 months ago

Good project at this age! Keep going!

taklaxbr5• 7 months ago

Thank you! I'm learning a lot from this process and the open-source community. Staying up late coding is definitely worth it when I see feedback like this! :)

jimmydin7• 7 months ago

nice! since you are a teen, you should check out Hack Club. Its a community of more than 100k teenagers, and funds teen's projects with donations from people (including elon!)

taklaxbr5• 7 months ago

Thanks a lot for the recommendation! I hadn’t heard about this before. Since my goal is to turn my projects into real products I’ll definitely check it out today.

Sabinus• 7 months ago

I find your 'including Elon' comment interesting. In my experience that would make people less likely to want to use the associated service. Is Elon's reputation different with the tech youth than the general public these days?

charlesding20242• 7 months ago

I dug into the repo – the 'self-healing' logic (automatically switching from CMD to PowerShell on failure) is genuinely clever. Much smarter than just dumping a stack trace.

A technical question on the v6 offline support (Phi-2): Does the CLI keep the model loaded in RAM in the background (daemon mode), or does it have to load the weights from disk every time an error occurs? I'm curious about the latency trade-off for that 'instant fix' feel.

taklaxbr5• 7 months ago

Thanks for the kind words regarding the self-healing logic!

To answer your question about v6/Phi-2: It uses a session-based RAM residency approach rather than a background daemon or per-request loading.

When you toggle the offline mode (or if it starts in that mode), the OfflineModelManager class loads the weights into memory once. Since the shell runs in a continuous while True loop, the model stays 'hot' in RAM for the duration of that session.

This eliminates the cold-start latency for every error correction, making the 'self-healing' feel instantaneous. The trade-off is, of course, the sustained RAM usage while the shell is open, but I found this preferable to waiting 10+ seconds for a re-load on every command failure.

charlesding20242• 7 months ago

That makes total sense. In a shell environment, breaking the flow for 10+ seconds would definitely be more painful than the memory overhead. The 'instant' feel is crucial for UX here. Thanks for the detailed explanation!

Show HN: I'm 15. I built an offline AI Terminal Agent that fixes errors

Related Stories