November 29, 2025
Eye spy with my DIY
Hachi: An Image Search Engine
Hachi wants to be your private photo detective — fans cheer, skeptics cry ‘AI hallucinations’
TLDR: Hachi is a self-hosted image search tool aiming to organize your personal photos privately across devices. The community loves the DIY privacy angle but clashes over reliability: fans say small local AI works great, skeptics warn it will lag and hallucinate as tech evolves. It matters for truly private search.
Meet Hachi, a DIY, self-hosted image search that promises to comb through your personal pics across devices like a faithful digital bloodhound. The author says it’s lean (just three dependencies!) and dreams of expanding to video, text, and audio—all while keeping your stuff private. The crowd? A glorious split-screen of hype and side-eye. One power user flexes that you can cobble this up with a tiny image-understanding AI, bragging they can “throw a hard drive at it” and get instant, local results. Others are just thrilled someone’s finally building a real personal search because, as one puts it, Windows and Mac “really don’t” do it well. Link for the curious: github.com/eagledot/hachi.
Then the drama kicks in. A thoughtful skeptic warns that search alone isn’t enough, arguing that large language models (text AIs) can be volatile and that we need better ways to keep track of where things live—call it digital “object permanence.” The spiciest take: a DIYer praises the engineering but says self-hosted AI will constantly fall behind because the tech that turns pictures into numbers (embeddings) evolves too fast—and points to a “girl drinking water” result as a clear hallucination. Cue memes about a cluster of refurbished smartphones becoming a Franken-search engine. Hachi is the new battleground where privacy-loving tinkerers clash with pragmatists who just want something that doesn’t get weird when asked to find a glass of water.
Key Points
- •Hachi is a self-hosted search engine intended for personal data distributed across devices and cloud services.
- •The current implementation supports image search, with plans to add video, text, and audio modalities.
- •The interface aims to expose multiple resource attributes and enable iterative, user-driven query refinement.
- •Semantic search is powered by self-hosted machine learning models accessed through a single interface.
- •Future goals include executing distributed queries across clusters of smartphones or single-board computers while maintaining performance.