Patrick Campanale has been in the tech space for well over a decade, that focus on PC/gaming recordsdata and opinions, as well to maker-focused merchandise to sign tiny companies.
With a open in expertise help in 2010 surrounding the Palm/webOS ecosystem, Patrick spent his childhood increasing cell functions as well to blogging for numerous publications, at final main to beginning his enjoy web insist in 2014. After working a expertise blog for a few years, he stepped out of that characteristic and into the area of excessive-quit custom-made PC manufacturing and building, with some degree of curiosity on YouTube video production and overclocking. Then, in 2017, Patrick joined the 9to5Toys crew as an editor/creator/reviewer with over 14,000 articles being published all the blueprint thru his time there, ranging from affords and roundups to in-depth opinions on basically the most up-to-the-minute expertise, video games, 3D printers, and extra.
In his free time, Patrick loves to salvage initiatives from wooden the use of assorted robots and programs, including leveraging the technologies of CNCs and lasers. If Patrick is now now not if truth be told working on a computer or taking half in video games, he’s possible in his 2-automobile garage workshop creating one thing lively. As well to all this, Patrick is also the media director at his native church where he feels God has known as him to abet, and he loves every minute of it.
I already sign all of my writing inner of Obsidian, so why now now not effect it so AI can search it from any tool I use? That’s what I’ve performed with this straight forward Raspberry Pi mission that upright needs a Pi, a tiny native model, and an Obsidian vault synced to it.
Obsidian LiveSync can whisk on one thing—Raspberry Pi included
It holds all of my notes to search thru with ease
I sign all of my writing in Obsidian—own for moderately a whereas now. I’ve gone thru the use of assorted apps to write over the past several years, including the use of Ulysses on Mac, Google Clinical doctors, Idea, Asana, and now Obsidian.
I delight in Obsidian on memoir of it retains the files native on my computer and in easy textual issue so they would maybe also additionally be read by one thing, whether or now now not that’s me or an AI system. I also delight in that I don’t must pay for a sync carrier or use a 3rd-birthday party platform to sync my notes delight in Google Pressure.
With Obsidian, I will use the initiate provide LiveSync plugin and own it sync my notes to the cloud of my preference—my enjoy system. This is a microscopic tricky, however certainly unruffled attainable.
I open by working CouchDB in Docker on the Pi, which is the provide of the LiveSync server. This retains the notes in an encrypted blob though, so they effect now now not appear to be readable appropriate off the bat.
Then, I whisk a headless version of Obsidian LiveSync on my Raspberry Pi, which syncs the notes into human-readable Markdown files, which is how I’m in a predicament to own my native AI items browse my notes.
Of route, I sign own offsite backups configured thru Duplicati as encrypted blobs to Google Pressure to effect sure the safety of my writings, however I delight in that all the pieces upright lives in the neighborhood on my computers and make contact with and syncs thru my enjoy hardware.
The Pi 5 can no doubt whisk some AI items on-tool
They gained’t compete with Claude or ChatGPT, however they’re free to whisk
Running native AI items on a Raspberry Pi is that if truth be told one thing that most efficient the Pi 5 can sign. You technically can whisk a local model on any Pi, nonetheless it upright gained’t be all that immediate and you’ll be moderately restricted except you own a first price quantity of RAM on no matter Pi you’re the use of.
However, for those with a Pi 5 (and especially would maybe own to you add the AI HAT+ 2 module), it’s no doubt attainable to whisk one thing delight in Qwen3 4B Deliver 2507 in the neighborhood at a first price drag. While it’s now now not frontier-grade, and even if truth be told all that mountainous of a model in any admire, it is moral at processing textual issue and surfacing recordsdata that you need to grab.
It’s moderately easy to run up Qwen3 4B Deliver 2507 in one thing delight in Ollama or llama.cpp (or thru hailo-ollama would maybe own to you own the HAT) and whisk it thru Launch WebUI to own a elephantine chat interface in uncover to search thru the files stored on it. I’ve extinct this several events to search out a say that I couldn’t get, or to help parse a few of the textual issue that’s in my vault.
All I own to sign is build an voice to AI where one thing lives and it might maybe maybe maybe get it for me
It’s delight in a non-public assistant that has excellent recall
It’s no doubt insane to mediate that I will whisk a local AI model on a Raspberry Pi and it no doubt be commended. I wouldn’t own Qwen3 4B Deliver 2507 or other smaller items help with a full bunch tasks spherical the house or address coding initiatives, however for easy textual issue and doc search it’s no doubt moderately succesful.
At any time once I want to search out a file, I will merely use Launch WebUI the use of the native Qwen3 4B Deliver 2507 model to search all of the notes which would maybe maybe well be on the Raspberry Pi. The mountainous thing is, since all of the notes are stored in easy textual issue and accessible on native machines, I might maybe maybe well also even whisk a extra extremely effective model on my desktop or notebook computer and search Obsidian if I desired to.
However, it’s nice so as to upright use the Pi self-contained for the procuring. Every thing stays there and all I use my other devices for is writing. This permits me to write from my iPad, Chromebook, notebook computer, desktop, or phone and now now not alarm about whether or now now not they’ve caught up in syncing to each and every other. As prolonged as no matter I’m writing on makes it help to the Pi, which is working LiveSync, I will see for it the use of Qwen3 4B Deliver 2507.
Raspberry Pi AI HAT+
- Compatibilty
-
Raspberry Pi 5
The Raspberry Pi AI HAT+ is accessible in two versions: 13 TOPS and 26 TOPS. Both are designed to work with the Raspberry Pi 5 completely, bringing AI computing to your favorite single-board computer.
This took moderately of hacking to salvage working, however my trouble is your sign
Really getting Launch WebUI to search thru an Obsidian vault is now now not if truth be told basically the most attention-grabbing thing to shut. It no doubt took me an ethical few hours to salvage it working successfully, including searching for to establish basically the most efficient model to make use of and one of many top ways it is possible you’ll be in a predicament to effect it no doubt obey the tool calls since low-parameter items don’t appear to be the smartest.
So, I’ve build the overall code that you need to sign this yourself on my GitHub to build you the voice that I went thru of getting this all to work. It’ll unruffled be moderately one-shot for you if you own the repo, however I’d certainly be sure that to examine it a few events to effect sure it if truth be told works the system you are trying it to. These tiny items sign desire to hallucinate till they’re precipitated into submission.
