3.10 Security & Data Privacy

From the very beginning, Pop was designed with privacy protection, local-first processing, and secure controllability as core principles.
Whether you are using local models, cloud models, knowledge bases, plugins, or workflows, Pop ensures transparency, safety, and user control over all data processing.

This chapter explains Pop’s security mechanisms and privacy protection strategies in detail.

🔐 1. Local First

Pop stores all data locally by default, including:

Chat history
Knowledge base documents and vectors
Workflow configurations
Plugin and tool configurations
Document editing content

Unless you explicitly enable a cloud model, no data is uploaded anywhere.

🔑 2. Model API Keys Stored Locally

All model API Keys (OpenAI, DeepSeek, self-hosted models, etc.) are:

Stored locally or in encrypted storage
Never uploaded to any server
Used only for model requests you manually trigger

You may delete, reset, or replace API Keys anytime.

🔏 3. Full Control Over Model Request Content

When using cloud models (GPT, DeepSeek, etc.), Pop sends only the following:

Your text input
Files you explicitly upload (images, PDFs, etc.)
Necessary model parameters (temperature, max_tokens, etc.)

Pop will not send:

Your entire local knowledge base
Other sessions
Unrelated private data

You control exactly what gets sent.

🛡️ 4. Tool Permission Control (MCP / Plugins)

Tools may involve accessing local files or executing commands, so Pop enforces strict permissions:

Tools are disabled by default (manual enable required)
High-risk tools (Shell, file writing) require explicit confirmation
All tool operations are auditable with visible execution logs
Tools follow the principle of least privilege

This ensures AI cannot access your system without permission.

🔍 5. Knowledge Base Security

All knowledge base content is stored locally:

Documents are not uploaded to external models
Vector indexes are saved in local storage
Chunking, slicing, and summarization are performed locally

During knowledge base Q&A:

Pop only sends small, relevant text fragments to the model
All other content remains local and inaccessible

Ideal for enterprise documents, private notes, internal architecture files, and sensitive data.

🧊 6. Multi-Window Isolation

Each chat window maintains its own independent context:

Window A’s messages do not affect Window B
Switching roles does not inherit other window contexts
Different models do not share conversation data

Perfect for handling sensitive tasks across separate contexts.

🗄 7. Local Data Access & Storage Location

Pop allows users to view local storage directories, including:

Chat sessions
User settings
Model configurations
Cache files
Knowledge base indexes

You may:

Clear caches
Delete sensitive data
Move knowledge bases to encrypted drives
Use private cloud or on-premise server mode (enterprise edition)

🧯 8. Privacy Mode (Future Expansion)

Future versions of Pop will include Privacy Mode, offering even stricter constraints:

Disable cloud model access
Prohibit file uploads
Disable all tool calls
Enable silent local-only mode

Suitable for enterprise, government, or highly sensitive environments.

🧾 9. Your Data Is NOT Used for Model Training

Anything you input into Pop:

Is not used for model training
Is never shared with other users
Is never provided to third-party datasets

This applies to:

Chat messages
Uploaded files
Documents
Knowledge base materials

You retain full ownership and control.

🧩 10. Enterprise / Team Advanced Security (Optional)

Enterprise deployment supports:

Local model servers
Intranet-only knowledge base indexing
Zero data exfiltration
Audit logs
Centralized key management (KMS)
RBAC permission control
Encrypted storage

Ensures corporate-level security, compliance, and isolation.

📌 Summary

Pop’s security system is built on three principles:

✔ Local First

All data is stored on your own device.

✔ Explicit Permissions

AI cannot perform system actions without your approval.

✔ User Control

All uploads, transfers, and tool calls are explicitly user-triggered.

This allows Pop to deliver powerful AI capabilities while ensuring your data always remains private, secure, and under your full control.