AI Transparency

Automatic Organization

When you save a note, AI automatically generates a title, tags, and a short summary so your ideas are easier to find later without manual filing. AI also suggests which folder a note belongs in based on your existing organization. These automatic features run every time you save content — you do not need to trigger them manually.

Semantic Search

Your content is converted into vector embeddings (numerical representations of meaning) so you can search by concept, not just keywords. You can find ideas even if you don't remember the exact words you used.

Voice Transcription

When you record a voice note in the iOS app, the audio is sent to OpenAI's Whisper service for speech-to-text transcription. The resulting text is then processed like any other note (titled, tagged, and stored).

Profile Learning

Magpie periodically analyzes your notes to build an understanding of your interests, active projects, and writing style. This profile is used to make AI suggestions more relevant to you over time — for example, generating tags that match your vocabulary or suggesting folders that fit your organizational habits.

Connection Discovery

AI may surface related notes or themes across your content to help you notice patterns or connections you might otherwise miss.

Optional Assistance

AI can help draft, summarize, or refine content when you ask for it through connected AI clients (see MCP section below). Your original words are preserved by default.

AI does not automatically rewrite or overwrite your content. AI-generated metadata (titles, tags, summaries) is stored separately and can be edited or replaced by you at any time.

AI does not publish, share, or distribute your content.

AI does not make decisions for you.

AI does not use your content for advertising or profiling.

OpenAI

Magpie uses OpenAI's API for content processing (GPT models for titles, tags, summaries, and folder suggestions), vector embeddings, and voice transcription (Whisper). Your content is sent to OpenAI's API over encrypted connections.

OpenAI's Data Practices

Under OpenAI's API data usage policy, API inputs and outputs are not used to train OpenAI's models. OpenAI may retain API inputs for up to 30 days for abuse and misuse monitoring, unless a zero-retention policy applies. For the most current information, see OpenAI's usage policies.

Note Processing

When you save a note, the note's text content (and URL or extracted page content, if applicable) is sent to OpenAI to generate a title, tags, summary, and folder suggestion. A sample of your existing tags and folder names may also be included for context so that suggestions are consistent with your existing organization.

Embeddings

Your note content is sent to OpenAI's embedding API to generate vector representations used for semantic search. The resulting embeddings (numerical arrays, not readable text) are stored in Magpie's database.

Voice Transcription

Audio recordings from the iOS app are sent to OpenAI's Whisper API. The audio file is transmitted over an encrypted connection. OpenAI returns the transcribed text. The audio file is not retained by Magpie after transcription.

Profile Learning

Periodically, a batch of your recent note titles, tags, content, and folder placements is sent to OpenAI to identify interests, projects, and writing patterns. The resulting profile summary is stored in Magpie and used to improve future AI suggestions.

Your Feedback Improves Suggestions

When you edit an AI-generated title, change AI-suggested tags, or move a note to a different folder than AI suggested, Magpie records that correction. These corrections are fed back into future AI prompts for your account so that suggestions improve over time and match your preferences.

Corrections Stay with You

AI corrections are stored in your account data. They are not shared with other users or used to influence suggestions for anyone else.

To avoid redundant processing and reduce costs, AI results (such as titles, tags, and summaries) may be cached for a period of time. If you save the same content twice, the cached result may be reused rather than making a new AI request. Cached results are stored in Magpie's infrastructure, not with the AI provider.

Your content is stored in Magpie's infrastructure (hosted on Supabase). If you connect GitHub, content is also synced to a repository you control as a backup.

AI-derived metadata (titles, tags, summaries, embeddings, profile insights) is stored alongside your content in Magpie's infrastructure. This data is part of your account and is deleted when you delete your account.

Magpie provides a Model Context Protocol (MCP) server that allows compatible third-party AI clients (such as Claude Desktop) to access your Magpie data. When you connect an AI client via MCP, that client can read and interact with your bookmarks, folders, and user context.

What This Means for Your Data

Data sent to third-party AI clients via MCP is processed according to that client's own terms, privacy policy, and AI practices — not Magpie's. For example, if you connect Claude Desktop, your data is processed by Anthropic under Anthropic's terms. Magpie does not control how third-party clients use, store, or process your data.

Your Control

MCP access requires explicit authorization through an OAuth flow. You choose which clients to connect and can revoke access at any time.

Not Used for Model Training

Your private content is not used to train AI models. OpenAI's API data usage policy states that API data is not used for model training.

Processing, Not Long-Term Retention

When AI processes your content, it does so to deliver a specific result (such as generating tags or transcribing audio). OpenAI may retain API inputs for up to 30 days for abuse monitoring. Magpie may cache AI results for a limited time to avoid redundant processing. Beyond these limited retention windows, your content is not retained by AI providers.

AI Isn't Perfect

AI may produce incomplete, inaccurate, or unhelpful results. Tags may miss the mark. Summaries may leave things out. Transcriptions may contain errors, especially with technical terms, accents, or background noise.

Use Your Judgment

Always review AI-generated output before relying on it for important decisions. AI is a support tool, not a source of truth.

Bias Awareness

Like all AI systems, the models used may reflect biases present in their training data.

You are responsible for how you use or share any AI-assisted output.

AI Amplifies, Not Replaces

AI exists to support your thinking, not replace your voice or creativity.

Sensible Defaults, Always Editable

AI runs automatically to reduce friction — you shouldn't have to manually tag and title every note. But every AI-generated result can be edited or overridden, and your corrections make future suggestions better.

Transparency Over Black Boxes

This page exists because you deserve to know exactly what AI does with your data, which provider processes it, and what the limits are. If something changes, I'll update this page and let you know.