1 Categories of Data Handled by midler

midler may handle the following categories of data, depending on which features you use:

• Authentication information — AI provider API keys; Enterprise Server username, password, and access tokens; email for password reset.
• Personal communications — chat prompts you submit to the AI, AI responses, and conversation history.
• User-provided content — uploaded or attached images, documents (PDF, DOCX, XLSX, PPTX, TXT, etc.), and pasted content.
• Website content — text extracted from the active browser tab when you explicitly invoke "Chat with page", "Summarize page", "Translate page", "Summarize near cursor", or related context-menu actions.
• Selected text — text you highlight on a web page, when you invoke "Search selection" / "Translate selection" via the context menu or keyboard shortcut.
• User preferences — UI language, theme, view mode, default LLM, prompt templates, policy rules, image generation settings, file format whitelist, etc.
• User memory — optional personal notes you save in "My Memory" to give the AI persistent context (e.g., "I am a software developer").
• Usage telemetry (only if Enterprise Server is configured) — feature type, LLM provider/model, response time, and excerpts of request/response text up to 5,000 characters per record.
• Internal AI ("RAG") queries (only if your organization enables it) — your search queries and selected service IDs sent to the organization-operated RAG endpoint.

midler does not collect: precise geolocation, financial information, health information, web browsing history (it does not record what pages you visit), or any data unrelated to features you actively invoke.

2 Data Stored Locally on Your Device

The following are stored on your device only and are never transmitted to us:

3 Data Sent to Third-Party AI Providers

When you submit a chat message, request a translation, generate an image, summarize a page, etc., midler sends the relevant input directly from your browser to the AI provider you have configured. midler does not proxy these requests through any server we control. The data sent depends on the action:

API keys are stored on your device and only attached to outgoing requests to the corresponding provider. Data is sent only when you explicitly initiate an action; midler performs no background data transmission.

4 Data Sent to Your Organization's Enterprise Server (Optional)

An Enterprise Server is opt-in: it is activated only if a server URL is configured (typically by an IT administrator) and the user logs in. When configured, midler transmits the following to that server, owned and operated by your organization (not by us):

• Authentication: server URL, username, password (during login over HTTPS), refresh tokens, password change requests, password reset email address.
• Profile sync: download of policies, LLM assignments, image generation defaults, allowed file formats, ACL/security settings, logging settings, branding, and other administrator-defined configuration.
• Usage telemetry: per AI request — feature type (chat/search/image/translation/file analysis/title generation/RAG/Gen-Design/Report-Gen), provider name, model name, response time, and a truncated excerpt of the request and response text (up to 5,000 characters each). The text excerpt is included so administrators can audit usage; this can include personal communications and uploaded content fragments. Logging can be configured by the administrator.
• User Memory: if you save personal notes in "My Memory", they are uploaded to the server so they can persist across devices.
• Policy submissions: any policy proposals you submit through the Extension's "Submit Policies" feature.

The Enterprise Server is operated by your organization. midler's developers do not receive or have access to data on that server. Please consult your organization's privacy notice for retention and access controls applicable to your account.

5 Internal AI ("사내 AI" / RAG) Search (Optional)

If the administrator enables the "Internal AI Search" feature, your search queries and selected service IDs are sent to the organization-operated RAG endpoint. Responses (including citations and document references) are returned to your browser and stored locally as part of your conversation. No data is sent to midler developers.

5b Code Interpreter (Optional, since v1.11)

The Extension includes an optional "Code Analysis" mode that lets the AI execute Python code locally in your browser using Pyodide (a WebAssembly Python runtime bundled with the Extension). Key points:

• Where code runs: Inside an isolated Web Worker in your browser. No code is sent to any midler-controlled server. All processing happens locally.
• What is sent to the AI provider: Your question, attached file metadata (file names and a system-prompt summary), and execution results (stdout text, error messages, return values). Chart images are not sent back to the AI; only a marker indicating a chart was rendered is included to save tokens.
• Attached files: When you attach a file in Code Analysis mode, the file's binary contents are mounted into Pyodide's in-memory virtual filesystem so Python can read them. The file is also written to your browser's IndexedDB for the duration of the conversation, then automatically cleaned up when the conversation is deleted or when the Extension is uninstalled.
• Generated code: The AI provider generates Python code via the same chat completion API. You can review every line of generated code in the collapsible "Python execution" block before/after it runs.
• External HTTP from Python: Generated Python may use pyodide.http.pyfetch to call external HTTP APIs. These calls originate from your browser, subject to browser CORS rules and the Extension's host_permissions. We do not record these calls.
• Charts: matplotlib charts produced by your code are stored locally in IndexedDB (midler_chart_results) so they persist across browser sessions. Deleting the conversation deletes the charts.
• Bundled analysis packages: pandas, numpy, matplotlib, openpyxl, and supporting libraries are bundled with the Extension. The Chrome Web Store build does not fetch additional packages from PyPI at runtime; all wheels load from the extension package.

6 Web Page and Selection Access

midler accesses the contents of web pages only when you explicitly invoke a feature that needs them:

• Chat with page / Summarize page / Translate page — extracts the visible text of the active tab and sends it to your configured AI provider.
• Summarize near cursor / Translate near cursor — extracts text near the right-click target.
• Search selection / Translate selection — uses the text you have highlighted.
• Analyze image (right-click) — sends the selected image and nearby text to your configured AI provider.
• Floating writing assistant — injects a small UI overlay onto the active page; text you type into it is sent to your configured AI provider.

Page contents are not stored centrally by us. They are sent to your configured AI provider, and the resulting AI response is stored only in your local conversation history.

7 Data Shared via Chrome Sync (Google)

Settings stored in chrome.storage.sync are encrypted by Chrome and synced across browsers signed in to the same Google account, per Google's Privacy Policy. We do not control this transport. If you do not wish settings to sync, you can sign out of Chrome Sync or disable extension sync in Chrome settings.

8 Data We Do NOT Share

• We do not sell, rent, or trade your data to advertisers or data brokers.
• We do not share data with analytics or tracking services.
• We do not use your data to train any machine learning model.
• midler's own developers do not receive any data from your installation; the Extension has no developer-controlled backend.

9 Permissions Requested

• storage — save settings and conversation history locally and via Chrome Sync.
• sidePanel — display the chat interface in Chrome's side panel.
• activeTab — read the current tab's text content for translation/summary/chat-with-page features.
• scripting — inject the writing assistant overlay and translation features into the page.
• contextMenus — provide right-click menu entries for selection-based AI actions.
• tabs — read the URL of the active tab to evaluate URL-based policies.
• host_permissions (all URLs) — required to inject the writing assistant on any page and to call AI provider APIs (which can be on user-specified domains).

10 Data Security

• All network calls to AI providers and to the Enterprise Server use HTTPS.
• API keys are stored in chrome.storage.sync and are not transmitted to any party other than the matching AI provider.
• We follow the principle of least privilege: data is sent only to the destination of the feature you invoked.
• You are responsible for the security of any API keys or credentials you enter into the Extension.

11 Data Retention and User Rights

Locally stored data persists until you delete it or uninstall the Extension. You may at any time:

• Delete individual conversations or all unpinned conversations from the chat history panel.
• Clear images, files, and gallery items.
• Reset all settings from the Extension's Settings page.
• Sign out of an Enterprise Server (your local cache of profile data is removed).
• Use the built-in Backup/Restore feature to export your data to a single .midler-backup file or to import one.
• Uninstall the Extension to remove all locally stored data including IndexedDB images.

Data stored on an Enterprise Server is governed by your organization's retention policy; please contact your administrator to exercise rights regarding that data (access, correction, deletion, export).

12 Children's Privacy

midler is not directed to children under 13 and does not knowingly collect data from children. If you believe a child has provided data through midler, please contact us so we can advise on local deletion.

13 International Data Transfers

When you use third-party AI providers, your data may be processed in countries other than where you are located, in accordance with each provider's privacy policy. Please review the linked policies in Section 3.

14 Changes to This Policy

We may update this Privacy Policy as the Extension evolves. Material changes will be reflected in the "Last updated" date and, where appropriate, surfaced in-product. Continued use of the Extension after an update constitutes acceptance of the revised policy.

15 Contact

For questions, complaints, or data-related requests, please contact us through one of the following channels:

• Contact form on our website
• GitHub Issues

Enterprise users should contact their organization's administrator for matters concerning data stored on the Enterprise Server.