Hey HN,
I'm Ravi, and I've spent the last few months building *SysNav* (
https://sysnav.ai). It's an intelligent terminal workspace designed specifically for SREs who want the speed of a native shell but the
context-awareness of an AI when things break.
**The Problem**
I found myself constantly context-switching: run `kubectl`, see error, copy to LLM, sanitize PII manually, paste, get generic fix, modify for my variables... it was friction-heavy.
**The Solution: SysNav**
I built SysNav as a "Assistant for DevOps and Ops". It wraps your terminal in a native "Intelligent Cockpit" interface:
1. **Ask Mode (Safe)**: You ask "Why is the redis pod crashing?". SysNav inspects the screen output (using a local PTY buffer), anonymizes PII, and sends *only* the context to the LLM. It returns a
suggested command. **It creates a human-in-the-loop safety barrier.**
2. **Agent Mode (Autonomous)**: If you trust it, you can give it a goal ("Find all zombie processes and kill them"), and it will execute the steps one by one, stopping if it encounters an unexpected
error.
**Security Architecture (The "Paranoia" Check)**
This was the most important part for me. SysNav uses a **Local-First Architecture**:
* **SSH Keys & Env Vars**: Never leave your machine. They are stored in your release of the Electron app.
* **Context**: We only send the text buffer of the terminal (logs/errors) to the inference layer. We have a redact-pii step that runs locally before network requests.
**Status**
We are currently in **Public Beta**. The core terminal is "fast-enough" (<100ms startup), and the AI context awareness is saving me about 30 minutes per incident.
I'd love for you to roast the architecture or tell me what features would make you actually trust an AI in your production environment.
Check it out: https://sysnav.ai
(Download available for macOS/Linux/Windows)
