YozoDeck Privacy Policy

Last updated: 14 October 2026

Yozoneko (“the developer”) provides this privacy policy for the desktop application “YozoDeck” (“the app”).

1. Overview

2. What the developer collects

Nothing.

• The app contains no analytics and no crash reporting, and no third-party library for either is bundled with it
• There is no sign-up, sign-in or account of any kind
• No code path exists that sends the user's information to the developer

3. What is stored on the user's computer

All of the following stay on the user's computer and are never transmitted.

• Settings: theme, font size, default shell, language and similar preferences
• History: the pane layout and a record of the work done with agents, kept so that a session can be restored. It is limited by age and size (180 days and 1 GB by default, adjustable in Settings); entries beyond the limit are deleted, oldest first
• Audit log: what pane-to-pane access, actions from a phone and answers to permission requests did, and when
• Rewind snapshots (Pro): copies of the working folder taken immediately before an agent changes it (a copy of the folder's contents, leaving out what .gitignore excludes and files such as keys and .env)
• Copies kept when resolving a collision: the file as it was before the app rewrote it
• Local AI data (if enabled): the Ollama runtime and AI models, the titles written for history entries, and the index used to search history
• Usage figures: totals computed from the logs each agent already keeps on the machine

These live under the user's own profile directory. Uninstalling the app and removing that data leaves nothing behind, with two exceptions that an uninstall does not remove: downloaded AI models (Ollama's model folder, ~/.ollama) and the entries added to ~/.claude/settings.json if the Claude Code integration was turned on. Turning the integration off in settings restores that file, and models can be deleted from settings.

4. Reading agent logs

• To show agent state, model and usage, the app reads the log and configuration files each agent keeps on the machine (~/.claude, ~/.codex, ~/.gemini and similar)
• What it reads is used only for display and local aggregation, and never leaves the machine
• If Claude Code integration (hooks and the status line) is enabled, the app adds settings to ~/.claude/settings.json. The exact text to be written is shown in full beforehand and written only if the user accepts it, and it can be undone at any time from the settings screen

5. Viewing from a phone

• When this feature is enabled, the app starts a small listener on the user's own computer so that it can be reached from the same local network, or from the user's own Tailscale network
• No server or relay operated by the developer is involved. The connection is directly between the user's computer and the user's phone
• The pairing secret that authorises a device is stored only on the user's computer, and access can be revoked at any time
• Traffic inside the same local network is not encrypted (plain HTTP). Someone else on that network could read the screen contents or the secret that authorises a device. Enable the feature only on a network you trust, and use Tailscale elsewhere (traffic inside Tailscale is encrypted)
• The feature is off by default and runs only when the user starts it

6. Local AI (generative AI)

• The app has an optional feature that runs a generative AI model on the user's own computer, using Ollama. It is off by default
• When enabled, pane screen contents, the instructions the user typed and similar text are passed to the model running on the user's computer in order to title history entries, summarise panes, explain permission requests and diffs, search history, and chat. None of this leaves the machine
• The exception is an endpoint the user enters in the settings: text is then sent to that endpoint (the Ollama instance the user chose)
• When the feature is enabled, and when the user fetches a model, the Ollama runtime is downloaded from GitHub (public releases of github.com/ollama/ollama) and AI models from the model distribution service operated by Ollama. These are ordinary download requests and carry none of the user's work. The size and location are shown before anything is downloaded
• AI output can be wrong. The app uses it only for explanations and search, never to decide an approval or to instruct an agent
• To report a problem with AI output, use the contact address below

7. Network connections made by the app

• Buying Pro (an in-app add-on) and checking whether it has been bought go through the Microsoft Store. Purchase information is handled by Microsoft; the developer receives neither the user's name nor payment details
• For the Microsoft Store edition, updates are handled by the Microsoft Store. The app itself contacts no server operated by the developer
• Copies distributed outside the Microsoft Store may fetch https://yozoneko.com/yozodeck/latest.json to see whether an update exists. Such a request carries nothing beyond an ordinary HTTP request and identifies no user

8. Programs running inside panes

The app is a terminal, and the programs running inside its panes — shells, AI agents and anything else — are started by the user. Whatever those programs send is outside the app's control. Please refer to the policy of each program's own provider.

9. Contact

• Provider: Yozoneko
• Email: [email protected]

10. Changes

This policy may be revised in response to changes in law or in the app's functionality. Significant changes will be announced on this page.