Android App — Personalization, Editable Memory & Native RAG Knowledge Base

Feature Request: Android App — Personalization, Editable Memory & Native RAG Knowledge Base

Summary

I would like to request three features for the DeepSeek Android app:

1. Personalization via custom system prompts / personas.
2. User-editable long-term memory management.
3. Native RAG (Retrieval-Augmented Generation) knowledge base with multi-format document support.

Background

As a long-term Android user, I find the app lacks the flexibility available on the web or through third-party tools. Specifically:

• No way to customize the AI's tone, persona, or domain focus via system prompts.
• No user-accessible long-term memory that can be viewed, edited, or deleted.
• No native support for personal knowledge bases, forcing users to rely on complex self-hosted solutions like Termux + Ollama + Open WebUI.

These limitations significantly reduce productivity in code assistance, academic research, and professional document Q&A scenarios.

Feature Requests

1. Personalization

• Add a "Personalization" or "Advanced Settings" section.
• Allow a global system prompt to define persona, response style, domain expertise, and language preference.
• Support saving multiple presets and switching between them quickly.

2. Memory Management

• Add a "Memory" page showing what the AI has remembered about the user.
• Allow viewing, editing, deleting, adding, exporting, and importing memories.
• Enable cross-conversation long-term memory with full user control and an option to disable it entirely.

3. Native RAG Knowledge Base

• Support importing common formats: PDF, Markdown, TXT, DOCX, EPUB, HTML, CSV, JSON, etc. Ideally with an extensible parser system.
• Build a local vector index after parsing; retrieve relevant chunks during Q&A.
• Show citations / sources in answers for verification.
• Knowledge base management: create multiple libraries, add/remove documents, re-index.
• Privacy mode: fully local parsing and retrieval (no uploads), or optional encrypted cloud sync.
• Reference existing open-source solutions like ToolNeuron and AnythingLLM, but integrate natively without requiring Termux.

Suggested Phased Implementation

• Phase 1: Local RAG for plain text / Markdown / TXT.
• Phase 2: PDF, DOCX, EPUB, and other common formats.
• Phase 3: OCR, audio/video transcription, and other extended formats.

Use Cases

• Personal notes, research papers, project docs, code repositories, e-books.
• Legal, medical, technical, and other professional domains requiring answers based on private documents.
• Offline or low-network environments where a local knowledge base is essential.

Expected Value

• Boost Android productivity for professional users.
• Enhance privacy protection and user trust.
• Reduce user churn to third-party self-hosted solutions.

Alternatives

Currently, users must rely on Termux + Ollama + Open WebUI or similar self-built setups, which are complex and fragmented.

Additional Context

I am willing to participate in beta testing and provide feedback. I can also provide screenshots or screen recordings to illustrate the specific scenarios.

Thank you for your continued work. I look forward to seeing DeepSeek improve.