Bit of an odd one for this forum but I think it's the right crowd. I had a box sitting here with a GPU that does basically nothing 90% of the time, already had a local model loaded on it for random stuff, and I was also in the middle of a miserable job hunt. Got tired of the routine of copying a listing into a chatbot, pasting my resume in after it, asking "am I a fit", then rewriting the whole application by hand. Did that about fifty times before deciding it'd be a better use of my evening to build something to do it for me. Naturally it took way longer than just doing the applications would have, but here we are.
github.com
Anyway the bit that's relevant to STH is that all the matching and assessment runs against a local OpenAI-compatible endpoint. So LM Studio / Ollama / llama.cpp / vLLM, whatever you've already got serving. You just give it the /v1 url and the model name and it runs everything on your own hardware. No cloud needed for any of it, no account, nothing phones home. Your resume and history just sit in a local sqlite db.
It's reasonably sensible about how it uses the model too, which matters if you're not running anything huge. It does a quick cheap pass over everything first and only spends proper tokens doing the deeper analysis (fit, where you're weak, how to pitch yourself for that specific role) on the ones that get through the first filter. Works fine on a fairly modest local setup, more context helps but it's not demanding.
There's an optional cloud hook (openai/claude/gemini) but it's purely for the "write the actual application in my own writing style" step, and only if you turn it on. I mostly use gemini for that because it's cheap, but you can leave it off entirely and the whole thing stays local.
Other stuff it does, briefly:
It's free and open source. Fair warning it's a proper desktop app with a bit of setup, not a website, so you do need to be comfortable installing things and pointing it at a model, but nobody here's going to struggle with that.
I keep adding bits and posting updates as I go, ko-fi's here if you want to follow the dev log or chuck a coffee at it: https://ko-fi.com/keljian . Works completely fine without, that's just where I dump changelogs.
GitHub - Keljian/JSE: Local-first desktop assistant for the whole job hunt — scrape listings, match with a local LLM, generate applications in your own voice, and track everything on a Kanban board.
Local-first desktop assistant for the whole job hunt — scrape listings, match with a local LLM, generate applications in your own voice, and track everything on a Kanban board. - Keljian/JSE
github.com
Anyway the bit that's relevant to STH is that all the matching and assessment runs against a local OpenAI-compatible endpoint. So LM Studio / Ollama / llama.cpp / vLLM, whatever you've already got serving. You just give it the /v1 url and the model name and it runs everything on your own hardware. No cloud needed for any of it, no account, nothing phones home. Your resume and history just sit in a local sqlite db.
It's reasonably sensible about how it uses the model too, which matters if you're not running anything huge. It does a quick cheap pass over everything first and only spends proper tokens doing the deeper analysis (fit, where you're weak, how to pitch yourself for that specific role) on the ones that get through the first filter. Works fine on a fairly modest local setup, more context helps but it's not demanding.
There's an optional cloud hook (openai/claude/gemini) but it's purely for the "write the actual application in my own writing style" step, and only if you turn it on. I mostly use gemini for that because it's cheap, but you can leave it off entirely and the whole thing stays local.
Other stuff it does, briefly:
- run multiple "lanes" at once if you're applying across a few different fields
- builds up a little knowledge base from your old applications so it actually learns your background and writes in your voice instead of that generic chatgpt sludge
- kanban board for tracking everything, interviews, feedback, follow ups, and it auto-archives stuff that's gone cold
- some market analysis, eg which companies/recruiters are flooding your area with matching roles and where your resume's got gaps vs what's actually being asked for
It's free and open source. Fair warning it's a proper desktop app with a bit of setup, not a website, so you do need to be comfortable installing things and pointing it at a model, but nobody here's going to struggle with that.
I keep adding bits and posting updates as I go, ko-fi's here if you want to follow the dev log or chuck a coffee at it: https://ko-fi.com/keljian . Works completely fine without, that's just where I dump changelogs.
