Mpesa Analyzer synchronizes your mobile messages securely and processes them using a local Large Language Model (Qwen 2.5 via Llama.cpp). Our system semantic-classifies financial senders across M-Pesa, bank notifications, SACCOs, and utilities without brittle regex patterns that break when SMS templates update.
From raw encrypted SMS to structured ledger, charts, and conversational AI insights.
Traditional budget trackers rely on manual entry or rigid text parsers that break whenever Safaricom, banks, or utility providers change wording.
Powered by quantized local LLMs running via Llama.cpp with CPU AVX2 and GPU support. The model extracts amounts, fees, counterparties, balances, and categories accurately.
Your financial data remains strictly self-hosted in your Docker deployment. Tokens use SHA-256 hashing, loot payloads are AES-256 encrypted, and zero data is sent to cloud APIs.
Sub-millisecond Redis 7 in-memory acceleration under peak load, with automatic, transparent fallback to MySQL sessions and disk cache if Redis restarts. Zero 500 errors, zero dropped logins.
Built-in real-time telemetry monitors memory RSS headroom, CPU load, MySQL connection pools, slow queries, and active LLM footprint with pure-SVG sparklines for zero front-end overhead.
An end-to-end ecosystem engineered for automation, reliability, and precision.
Ask questions about your transactions in plain English or Sheng. The assistant automatically injects your real spending aggregates, Fuliza fees, recipient breakdowns, and budget goals to deliver instant personalized advice.
Deploy your private instance or start managing your transactions with local AI precision.