AI · Python
CH3SH1RE is a privacy-first AI assistant that runs entirely on my local hardware, built as a direct response to the data exposure that comes with cloud-based AI tools. Written in Python with SQLite for persistent memory, it routes queries to a locally-hosted language model meaning no API calls, no external logging, and no dependency on third-party servers. The project started from a single principle: that a personal AI assistant should be genuinely personal. Although I can't say much because my own hardware limits me.
Runs Qwen2.5 7B instruct (q5_k_m quantized) entirely on-device via a llama.cpp server. No internet required after setup.
doneConversations and context stored in a local SQLite database, letting the assistant remember past interactions.
doneImage understanding through a dedicated vision service, allowing the assistant to process and describe visual input.
in progressPlanned integration to give the assistant access to live web results to answer questions beyond its training data.
plannedPlanned support for loading personal documents and files so the assistant can answer questions grounded in your own data.
plannedExploring function calling and tool-use patterns to let the assistant take actions, not just answer questions.
plannedqwen2.5-7b-instruct-q5_k_m.gguf