Dynozen is a desktop SSH client with an AI agent that investigates your servers: it runs read-only diagnostics first, explains what it found, and pauses for your approval before anything changes state. Everything stored locally.
Ask the question you actually have, like why is nginx returning 502?, and it runs the diagnosis itself: unit status, error logs, the journal. Read-only first, every time, streamed into your terminal.
Anything that changes state (restart, kill, rm, anything under sudo) pauses behind one card: the purpose and the exact command. Nothing runs without your word.
A denied command is never retried and never rephrased into a second attempt. The agent stands down and stays read-only.
After an approved fix the agent re-runs the same reads that found the problem: unit state first, then the failing request itself.
Every session leaves a log: each command, who ran it, read-only or approved. No summary to trust, just a log to read.
The agent sees the same screen you do. Ask about the output in front of you, with no copy-pasting production logs into a chatbot tab.
CONTEXT · CURRENT TERMINALOpenAI, Anthropic, Google Gemini, or any OpenAI-compatible endpoint. Your API key, called directly from your Mac, with no middleman between your servers and your model.
The parts every SSH client has: present, working, not the story.
Not a promise, but a transcript. Checkout went down at 03:07; here is the whole investigation, exactly as the agent panel shows it. The Approve button is yours.
Not a badge wall. The actual permission boundary, in one table.
~/.ssh, stored in a local vault (AES-256-GCM, Argon2). No account, no cloud, nothing leaves this Mac. Host fingerprints are pinned on first connect.xattr -dr com.apple.quarantine /Applications/Dynozen.app