Status: beta, a solo project under active development. Issues and feedback are welcome; there are no formal support or stability guarantees yet.
A personal, local-first job-search dashboard: track target companies, pull their openings directly from applicant-tracking-system job boards, score every posting against your own written fit criteria (optionally with AI, using your own Anthropic API key), and manage applications, contacts, and reminders, all on your own machine.
Setup is the application. A fresh install boots into a wizard that is the whole app (no empty dashboard to bounce off of) and leads with the privacy promise.
The scoring reader shows its work. Every posting is scored against the wish-list criteria you ranked, and each criterion is mapped to evidence from the actual posting. Every score shows its reasoning.
The tracker never fakes progress. A persistent "Setup N of N" pill tracks real completeness, fed verbatim by the backend's own count.
Why these choices? The senior decisions here are design and systems judgment carried into code: a setup flow designed as reflection rather than a form, a token architecture that catches its own theme gaps, scoring that lives in a document you write, and a color-blind-safe state model with a simulator guarding it. The case study walks through each and names the things deliberately left unbuilt.
- Local only. One process on localhost. Your data lives in a SQLite file you own. No accounts, no server, no sync.
- Zero tracking. No telemetry, no crash reporting, no update pings, no install beacons. The app talks to job-board services for the companies you add (finding and pulling their openings, plus a per-company logo lookup) and to the AI provider you configure: the Anthropic API with your own key, or any OpenAI-compatible endpoint you point it at, local runtimes like Ollama included (AI features are optional; the app works without either). PRIVACY.md is the complete inventory of every request.
- Cross-platform. Python + FastAPI + a framework-free web frontend. Mac, Windows, and Linux.
- Your criteria are the product. Scoring is driven by a criteria document you author: hard filters (comp, location, sector, seniority) and weighted, ranked criteria in your own words. You can read and change exactly what drives every score.
pip install .
jshq
Then open http://127.0.0.1:5747. First run creates a data directory
(~/Library/Application Support/jshq on macOS, %LOCALAPPDATA%\jshq on
Windows, ~/.local/share/jshq on Linux, or wherever JSHQ_DATA_DIR
points) seeded with editable copies of the example fit-criteria doc and the
voice guide.
Two commands keep an install current: jshq refresh runs one ATS ingestion
pass (meant to run twice a day), and jshq backup takes one verified backup
(a SQLite snapshot plus dated copies of your criteria doc, voice guide,
roadmap, resume content, and a mirror of uploaded application files) into
backups/ inside the data directory, keeping the newest 30 of each. The
.env holding your API key is deliberately not included; copy it somewhere
safe yourself.
You don't have to schedule either by hand:
jshq schedule --install
writes the native scheduler entries for both (launchd on macOS, cron on
Linux, Task Scheduler on Windows) with refresh at 10:00 and 16:00 and backup
at 02:00. Change the times with repeatable flags
(jshq schedule --install --refresh-time 08:00 --refresh-time 12:00 --refresh-time 18:00) or from Settings → System → Scheduling in the app;
--status shows what's installed and --uninstall removes it. Re-running
--install replaces the entries, never duplicates them. On a system without
a supported scheduler the command prints manual instructions instead: point
your own scheduler at jshq refresh twice a day and jshq backup nightly
(for example 0 10,16 * * * jshq refresh and 0 2 * * * jshq backup in
cron), with JSHQ_DATA_DIR set if you use a custom data directory.
The optional AI features run on a provider you choose in Settings →
System (the setup wizard offers the same choice): Anthropic with your own
API key, saved to a .env in your data directory on this machine and sent
only to api.anthropic.com, or any OpenAI-compatible endpoint you run or
trust (Ollama, LM Studio, a hosted provider) with a free-text model id.
Switching providers keeps both configurations saved, and an Advanced control
can route analysis and writing work to different providers and models. The
same screen sets your persona (the name the AI writes as, or none) and your
voice guide. Without a key or endpoint the app runs fully; AI features
simply stay off with an in-app note.
PDF rendering (resume/cover letters) uses an installed Chrome, Chromium,
or Edge; set JSHQ_CHROME if yours lives somewhere unusual.
python3.12 -m venv .venv
.venv/bin/pip install -e . --group dev
.venv/bin/pytest
Copy .env.example to .env and set JSHQ_DATA_DIR=./data to keep dev
data inside the checkout.
Built by Chris Hays. The repository is at github.com/noestudios/jshq.
The onboarding's two reflection exercises (the ranked wish list and the fulfillment matrix) are adapted from tier-list ranking and fulfillment-matrix exercises by Kristin Chen. Visit www.kristinmchen.com for her career coaching, startup advising, and fractional CPO services. Thank you, Kristin.
AGPL-3.0. See LICENSE.