Why Local AI Matters for Your Finances
The problem with cloud AI in finance apps
When finance apps categorize your transactions using cloud AI, your data travels to remote servers, gets processed by third-party AI models, and the results come back. Your entire financial history passes through infrastructure you don’t control.
This isn’t hypothetical concern. It’s how these services work by design.
How PocketVault does it differently
PocketVault runs AI entirely on your device:
- Transaction categorization uses a local LLM via llama.cpp
- Semantic search uses ONNX embeddings running on-device
- Natural language queries are processed locally with RAG
The AI models are downloaded once and run entirely offline. Your transactions never leave your device for AI processing.
Does local AI sacrifice quality?
Not at all. Local LLMs handle clear transactions with over 90% accuracy, and continue to improve on ambiguous ones. The key advantage is that this processing happens entirely on your device — no data leaves, no matter how many transactions you categorize.
The semantic search engine uses all-MiniLM-L6-v2, the same model used by many production systems. Search coffee and find Starbucks, Dunkin’, and your local cafe.
The future is local
Apple Intelligence proved that on-device AI is the future. PocketVault applies the same principle to your finances. Your data stays yours.