Privacy Policy
How Robin processes and stores information across your local Windows computer, AI models, and connected services.
1. About Robin
Robin is a desktop personal assistant application for Windows designed to help you organize daily information—including email, calendar events, tasks, contacts, files, and notes—through a unified conversational interface.
Robin operates with a local-first design: the application runtime, configuration, local database, and session records run on your physical Windows PC rather than on remote centralized servers operated by Robin. When you choose to connect third-party services or cloud AI models, information flows directly between your computer and those authorized providers according to your settings.
2. Information Robin Processes
Depending on which features you enable and use, Robin processes:
- User Requests: Text messages, spoken instructions, and commands you submit in conversation.
- Connected Account Data: Email headers, message content, calendar events, task items, and contact information retrieved via services you authorize (such as Google Workspace).
- Workspace Files: Content and metadata of documents located strictly within your configured local Robin workspace directory.
- Local Memory & Preferences: Saved context, project names, confirmed contact identities, and assistant preferences stored on your device.
- Microphone Audio: Live audio captured in memory by your microphone during active voice interaction.
3. Local Processing and Storage
Robin stores application data on your local hard drive in your Windows user profile under ~/.robin. This local storage consists of:
- Application Database (
robin.db): A local SQLite database storing confirmed contacts, action history, and configuration preferences. - Session Database (
adk-sessions.db): A local SQLite database storing conversation logs and turn history on your machine. - Embedded Vector Store (
qdrant/): A local embedded vector index used for contextual memory search on your device.
4. AI Model Providers
Robin allows you to choose how assistant reasoning is handled:
A. Local AI via Ollama
When configured to use local AI through Ollama (connecting to http://localhost:11434), model reasoning executes on your computer's hardware. Prompts and request context are not sent to cloud AI providers.
B. Cloud AI (Google Gemini)
When configured to use a cloud model such as Google Gemini, relevant user instructions, conversation context, and tool inputs are transmitted directly from your desktop to Google's Gemini API using your configured API key. That data is handled under Google's applicable terms and policies.
5. Google User Data
Robin offers optional integrations with Google Workspace. Robin accesses Google user data only after you explicitly initiate connection and authorize permissions through Google's standard OAuth sign-in flow.
Google API Services User Data Policy Compliance
Robin's use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements.
A. Exact Scopes & Purpose
Robin requests only the specific permissions necessary to deliver assistant functionality that you command:
https://www.googleapis.com/auth/gmail.readonly: Used to inspect message headers and read emails when you explicitly ask Robin about your inbox or request details from a specific email thread.https://www.googleapis.com/auth/gmail.compose: Used to prepare and draft email messages for your explicit visual review and confirmation before sending.https://www.googleapis.com/auth/gmail.modify: Used to update labels, mark messages as read, or modify messages strictly when directed by you.https://www.googleapis.com/auth/calendar.events: Used to read your schedule, check for scheduling conflicts, and draft new events or updates on your Google Calendar.https://www.googleapis.com/auth/tasks: Used to view, organize, create, and update tasks in your Google Tasks lists.https://www.googleapis.com/auth/contacts.readonly: Used to read contact names and email addresses to resolve who you mean when asking Robin to contact or reference a person.
B. How Google Data Is Handled
- Robin uses authorized Google data solely to deliver user-facing assistant features requested by the user.
- No generalized AI model training: Robin does NOT use Google Workspace user data to train, retrain, or fine-tune generalized machine learning or artificial intelligence models.
- Robin does not use Google user data to display, target, or serve advertisements.
- Robin does not sell, rent, or transfer Google user data to data brokers, advertising platforms, or commercial third parties.
- Actions that modify external state—such as sending an email or saving a calendar event—require your explicit human review and approval in Robin before execution.
C. AI Model Processing with Google Data
How Google data interacts with AI reasoning depends on your chosen model provider:
Local AI (Ollama)
When using local Ollama models, all reasoning takes place 100% on your local PC. No Google Workspace data or prompts are transmitted to any external cloud AI provider.
Cloud AI (Google Gemini)
When using Gemini, relevant excerpts from your request (e.g. an email snippet needed to draft a reply) are transmitted directly to the Google Gemini API using your personal API key over encrypted HTTPS.
D. Disconnecting Access & Credential Storage
Google OAuth tokens are stored exclusively in your local Windows Credential Manager under the service name Robin. You can disconnect your Google account at any time in Robin Settings, which deletes the tokens from Windows Credential Manager. You can also revoke access at any time through your Google Account permissions page.
6. Files and Workspace Access
Robin can read and summarize local documents you place in its designated workspace directory (defaulting to C:\RobinWorkspace or your customized folder).
Robin's file tools are scoped to this directory. Robin does not perform background scans of your entire hard drive, personal folders, or system files.
7. Voice Features
Robin includes voice capabilities using on-device speech processing:
- On-Device Transcription (Moonshine): Microphone audio is processed locally on your computer's CPU using the on-device Moonshine speech-to-text model. Spoken audio is converted to text locally before any assistant action occurs.
- Volatile In-Memory Processing: Audio frames reside strictly in volatile memory buffers during active speech processing. Robin does not write permanent audio recording files to disk and does not stream raw microphone audio to remote servers.
- User-Initiated Voice: In Robin V1, voice interactions are active and strictly initiated by the user. Robin processes speech only when deliberately activated and does not monitor background audio.
- Cloud AI Distinction: When you speak to Robin while configured with a cloud AI provider (such as Gemini), your voice is transcribed to text locally on your PC, and only the resulting text prompt is transmitted to the cloud model.
8. Memory and Personal Context
To maintain conversational continuity, Robin can remember relevant details you share (such as preferred meeting times, ongoing project names, or confirmed contact identities).
- Memory records and semantic embeddings are stored locally on your device in
robin.dband the local vector index. - Robin does not claim to remember everything; it retains specific context relevant to your daily tasks.
- Memory is not hidden surveillance—it is stored in your local application data folder.
9. Credentials and Connected Accounts
Robin does not store your Google or third-party account passwords. When you connect an account via OAuth, access and refresh tokens are stored locally in the standard Windows Credential Manager under the service name Robin using the operating system's built-in credential locker.
When you provide a Gemini API key in Robin Settings, Robin stores the key using Windows Credential Manager rather than in Robin's plaintext configuration files. Non-sensitive AI configuration, including the selected provider and model, is stored locally in ~/.robin/ai-settings.json.
Tokens and credentials are not stored in reviewable action plans, unencrypted application logs, or transmitted to Robin servers.
10. Human-in-the-Loop & Action Approvals
Robin distinguishes between reading information and modifying external state. Actions that change something—such as sending an email, creating or editing a calendar event, or updating a task—are prepared in a pending state and require your explicit review and approval before execution.
Pending actions that are not approved automatically expire after 24 hours in the local database.
11. Website Analytics & Desktop Telemetry
A. Public Website Analytics (robinpa.in)
We use Google Analytics 4 on our public website (robinpa.in) to understand general visitor traffic, measure engagement with website content, and improve user experience.
Google Analytics collects high-level site interaction information, such as:
- Page visits, route navigations, and timestamps
- Referral sources (e.g. search engines or direct links)
- Browser type, device model, operating system, and screen resolution
- Approximate geographic region (country and city level inferred by Google)
- Website interaction events (such as clicking the Robin download link or feature discovery CTAs)
Google processes this information under Google's Privacy Policy. You can prevent Google Analytics tracking across the web by installing the official Google Analytics Opt-out Browser Add-on.
B. No Personal Assistant or User Data Sent to Analytics
Robin does not intentionally send user assistant data to Google Analytics. Specifically, Google Analytics never receives:
- Your Robin conversations, prompts, or messages
- Gmail messages, drafts, or inbox data
- Google Calendar events, schedules, or meetings
- Google Tasks or lists
- Google Contacts or people records
- Workspace file names, contents, or directory paths
- Microphone audio recordings or voice transcripts
- Local semantic memories stored on your device
C. Desktop Application Telemetry
Google Analytics is deployed solely on the public robinpa.in website. The Robin Windows desktop application does not include Google Analytics, advertising SDKs, or background telemetry trackers. Robin does not monitor background computer activity, keystrokes, or browser history outside the assistant interface.
12. Data Sharing and Selling
Robin does not currently sell personal information to advertisers.
Information is shared only with third-party services that you explicitly configure or authorize (such as your chosen cloud AI provider or connected Google Workspace account) to carry out assistant requests.
13. Data Retention and Removal
Because Robin stores its application data locally on your computer, you can manage and remove your data directly:
- Robin Application Data: Deleting the local directory
~/.robinremoves Robin-managed files, including your local SQLite databases (robin.dbandadk-sessions.db), the local Qdrant vector index, local AI settings, and Robin-cached voice models (such as Moonshine). - Credentials and API Keys: Stored credentials—including Google OAuth tokens and your Gemini API key—are managed in the Windows Credential Manager. Disconnecting your account or clearing keys in Robin Settings removes those credentials from Windows Credential Manager.
- Ollama Models: If you use Ollama for local AI, models you download are managed and stored independently by Ollama on your system. Deleting the
~/.robindirectory does not delete your Ollama installation or downloaded Ollama models; these can be managed or deleted directly through Ollama.
14. Security
Robin uses standard security practices suitable for Windows desktop software:
- Storing account tokens in the Windows Credential Manager rather than plaintext files.
- Using standard OAuth 2.0 PKCE authentication flows.
- Restricting file operations to designated workspace paths.
- Requiring human approval before consequential actions execute.
No software system is entirely immune to security risks. You are responsible for maintaining the physical and operating-system security of your Windows PC.
15. Children's Privacy
Robin is intended for general productivity use by adults. It is not directed to children under 13, and we do not knowingly collect personal information from children.
16. Changes to This Policy & Contact
We may update this Privacy Policy as Robin develops. Any updates will be posted to this page with an updated revision date.
Contact
Official public contact mailboxes will be provided on this website prior to public distribution.