Privacy and data handling

Privacy Policy

Last updated October 9, 2026

This policy covers the Viscacha macOS application, viscacha.app, and its optional launch waitlist. Viscacha is provided by Nikolai Vladimirovich Solovev, an independent developer based in Japan ("we", "us", or "the developer"). For privacy questions or requests, contact legal@viscacha.app. Our postal address is available without undue delay on request at that address.

In current development builds, we do not collect or store your app prompts, selected text, image attachments, AI responses, request history, API keys, or proxy passwords on our servers. App data is handled on your Mac and sent to the AI provider or endpoint you choose when needed to carry out your request. Planned hosted access will use a different data flow, described below. The separate website waitlist stores the email address you voluntarily submit.

What the app does

Current development builds use AI models through your own provider accounts, API credentials, or local model servers. You choose what to submit and which provider, model, endpoint, and optional proxy to use. These builds do not provide a Viscacha user account, hosted AI inference, or cloud storage or synchronization of your app content.

The app does not include advertising trackers, developer usage analytics, or automatic crash-report uploads to us. We do not sell app content or use it to train AI models.

Data stored on your Mac

• Settings and configuration: preferences, keyboard shortcuts, saved prompt templates, provider and endpoint details, proxy addresses and usernames, and model preferences are stored locally in app preferences or files.
• Secrets: API keys and proxy passwords entered in Viscacha are stored separately in the macOS Keychain and retrieved to authenticate the connections you configure.
• Request history: history is enabled by default. It can include the selected text, prompt, question, AI response, available reasoning text, model, time, token counts, duration, and estimated provider cost. You can turn off saving future history in Settings and delete existing entries in History. Turning history off does not erase previous entries.
• Caches and temporary files: the app keeps a local model catalog cache. Image attachments used with Codex are temporarily written to local files; the app attempts to remove those files when the request finishes or is cancelled. An interruption or crash can leave temporary files behind.
• Local diagnostics: macOS app logs can contain operational details, errors, and short excerpts of selected or clipboard text. These logs are not automatically sent to us. Turning off request history does not disable system logs.

Local history and configuration are not protected in the same way as Keychain secrets. Their protection also depends on your Mac's account, disk encryption, backups, and any tools you use to copy or synchronize local files.

Selected text, clipboard, and permissions

When you invoke an app action, Viscacha can read selected text through macOS Accessibility. As a fallback, it can use Automation or a simulated copy command, then read the clipboard. If no selection is available, the action may use existing clipboard text or images, including local image files copied to the clipboard. You can also supply text, questions, and image attachments yourself.

These functions use the macOS permissions you grant for accessing selected content and files. You can revoke permissions in System Settings. The app does not continuously monitor your clipboard or record all keystrokes. Check the content and destination before running a request, especially if your clipboard contains confidential information.

AI providers and endpoints you choose

For a direct remote AI request in current development builds, your Mac sends the prompt, selected text or question, supported image attachments, model settings, and authentication information to the chosen provider or custom endpoint. Connections are made from your Mac, optionally through your configured proxy; these requests are not routed through a Viscacha AI backend. Refreshing a provider's available model list also contacts that provider or endpoint.

Supported integrations include Anthropic Claude, OpenAI, Google Gemini, xAI Grok, DeepSeek, Groq, compatible custom endpoints such as OpenRouter, and local servers such as Ollama, LM Studio, and llama.cpp. The recipient can receive your IP address and connection information as well as the request content. Its own privacy policy, retention rules, account settings, and terms govern its processing. We cannot access or delete data held in your provider account or guarantee that a provider will not retain it or use it for training.

Apple Foundation Models processes supported requests on your device. A local model server keeps inference on your device only if you configure an actually local endpoint and the server itself does not forward data elsewhere. Using a local model does not prevent the app's other connections described in this policy.

Planned hosted access

We plan to offer paid Viscacha Pro subscriptions with hosted Claude access and a usage cap, using API credentials managed by Viscacha. Users will not need their own API key for this service. Hosted access is not currently available. Request content, including the prompt and selected text or image, will pass through our backend to Anthropic, and responses will return through that backend. Our initial backend is planned for Japan, with a possible deployment in Germany. Claude access will be limited to countries supported by Anthropic.

Saved prompt templates and app history are intended to remain local to your Mac. Processing a hosted request still requires transmitting its content. We will not use that content to train AI models. Before offering hosted access, we will update this policy with the actual data processed, service providers, request routing, retention, and deletion rules, and provide any notice or consent required by law.

Codex integration

If you choose Codex, Viscacha uses a compatible Codex runtime installed on your Mac and your existing Codex authentication. Codex manages its own credentials; Viscacha does not copy them into its API-key store. The app can read account status, including an account email or plan type supplied by the runtime, to determine availability.

Requests are passed to that runtime and processed under the relevant OpenAI terms and privacy settings. Viscacha asks the runtime to create temporary, read-only sessions. That setting does not guarantee that OpenAI or the runtime keeps no diagnostics or other records. Viscacha's own local request history can still be saved if enabled.

Other network connections

Model catalog: the app downloads public model information from viscacha.app/models/models.json on startup and when refreshed. This request contains normal connection information, such as an IP address and HTTP headers, but does not upload your prompts, API keys, images, or history. Our hosting and network providers, including Cloudflare, can process technical request information to deliver and secure it.

Proxy checks: if you test a proxy or enable strict proxy verification, the app contacts ipify (api.ipify.org) directly and through the proxy to compare public IP addresses. ipify receives connection information, including the relevant public IP address; no AI prompt or API key is sent for this check. Your chosen proxy operator also processes the traffic carried through it.

External content: opening links or loading remote images contained in an AI response can contact the relevant website or image host. Its own privacy practices apply.

The app's Offline Mode controls which AI credentials may be used. It is not a guarantee that every app or runtime connection is blocked: the public model catalog can still be requested, and a credential you allow may point to a remote endpoint.

Exports, deletion, and your control

You can manage saved prompts, credentials, proxies, and history in the app. Deleting an individual credential or proxy also removes its associated secret from Viscacha's Keychain store. A bulk configuration reset should not be treated as erasing Keychain secrets or all local records.

If you export app data, the file is saved where you choose. Selecting the option to include sensitive data writes API keys and proxy passwords into that export in plain text. Export files, copied results, backups, system logs, caches, and separate Codex data remain under your control and may need to be removed separately. Uninstalling the app alone does not necessarily erase them. Deleting local records does not delete copies already received by an AI provider.

We cannot remotely retrieve, correct, or erase app content held only on your Mac. Use the app and macOS controls for that data, and contact your provider for data it holds.

The website and optional launch waitlist

Visiting the website or using the app does not enroll you in the waitlist. If you choose to join, we store your email address and signup time in a private database on our server in Japan. We use that address only to send exactly one email when Viscacha launches. We do not send a confirmation email, newsletter, promotions, or beta updates, sell the address, or add it to another mailing list.

Cloudflare delivers and protects the website and provides the Turnstile security check near the signup form. It processes technical information such as IP address and browser information. Our signup server also uses IP addresses temporarily in memory to limit abuse; the waitlist database stores only the email and signup time, and the signup API does not log submitted emails, IP addresses, or security tokens. We do not use advertising pixels or analytics trackers. Website images and fonts are served from our site.

Cloudflare's processing is described in its Privacy Policy and Turnstile Privacy Policy. Necessary delivery and security information may be processed outside Japan by our infrastructure providers. Providers you choose for AI requests may also process data outside Japan.

Waitlist access is restricted to the developer and authorized operators. The database is outside the public website directory and has no public list or export endpoint. If an email delivery provider is used for the launch message, we will identify it in this policy before transferring the list to it and address any applicable transfer requirements.

We keep your signup until the launch notification is sent, you request removal, or the project is discontinued, then delete it. The same purpose and deletion rules apply to retained waitlist exports and backups. To access, correct, or delete your signup or withdraw your launch-email request, email legal@viscacha.app, preferably from the address you used.

Messages you send us and privacy requests

If you email us, we receive your email address, message, and any attachments you choose to send, through our email provider. We use this information to answer you, handle your request, and meet applicable legal obligations, and keep it only as long as reasonably necessary for those purposes. Please do not send API keys, proxy passwords, or unrelated sensitive content.

You can request information about data we hold about you, access, correction, deletion, or restriction of processing where applicable law provides those rights. Contact legal@viscacha.app for these requests or complaints. We may ask for proportionate verification to avoid disclosing or deleting someone else's data. We do not charge for ordinary privacy requests. Nothing in this policy limits your rights under applicable law.

Changes

We will publish changes here with an updated date and provide additional notice when required by law. We will not repurpose waitlist addresses beyond the one launch notification you requested without an appropriate legal basis and any consent required by law.