Free and open source. MIT license, no subscription, no ads, no accounts.
ApplyPack finds real openings, shows exactly which words a posting wants and your resume lacks, helps you fix it in place, and writes the cover letter. On your own machine.
Type Redis into the skills line: the score moves on the next keystroke. Type hooks, a must-have, and it moves further. Before you type Terraform, its dashed underline in the posting says nothing in the resume backs it. Typed, it counts all the same: the score reads the text, as an ATS filter does. A word only in a list counts a little less than one in a sentence about the work. Title, summary and latest role keep the AI's grade, so rewrite most of the resume and the page stops showing a number.
The real app on a synthetic resume, with real AI calls and no cuts. Twenty-five seconds.
The Overview, the jobs it ranked, one posting's verdict, the resume compared with it, every suggestion applied in one press (85 to 88, no AI call), the resume as the document itself with the changes marked, the fact-checked letter, and one resume against three postings: 90, 85, and a straight 0 where the stack is PHP.
A resume that says "PHP 8, Laravel, Symfony" can miss a requirement that says "PHP", and a recruiter never sees it.
The wrong stack in paragraph four, "Remote" that means remote in Germany, a listing nobody will ever fill.
Refreshing boards, reading, guessing what to change, rewriting the letter. ApplyPack was built during my own search to take that over.
A blocked claim sends the draft back once with the violation quoted. Still blocked: nothing is stored.
Cheap deterministic filters drop the obvious misses first, so the AI only reads postings that might be for you.
Most AI resume tools average their way to a flattering number. ApplyPack splits the job: the model only marks what is present, missing or unverifiable, and a unit-tested formula applies the caps.
Off by default, one switch in Settings. Criteria in your own words, names removed before any model reads a word, every answer with its quote from the resume, the score in code, and a person who decides.
I was mid job hunt when I started ApplyPack. Every evening went to refreshing boards and reading postings that turned out to be the wrong stack, the wrong country, or nobody's real vacancy. So I built the tool I wanted and ran my own search through it every day. That is how I found the job I have now.
It is the same tool you install, unchanged.
Nazar Boyko, September 2026
ApplyPack is MIT-licensed and has no hosted version: no accounts, no subscription, no telemetry, no ads. You bring the AI: a subscription you already pay for, an API key, or a local model that costs nothing.
It is built to be changed. Over 2,600 unit tests on the pure modules, a decision record for every non-obvious choice, and a CLAUDE.md that says where everything lives, so a first change takes an evening, with an AI assistant or without one.
All you install is Node.js 22 or newer: the database comes with ApplyPack, no Docker needed. Step by step for macOS, Windows and Linux.
git clone https://github.com/applypack/applypack.git cd applypack npm install npm start # → http://127.0.0.1:4747, the setup opens
On a server, or to keep it running without a terminal: docker compose up -d runs the same thing in containers.
Paste an AI key on the first screen, or let the dashboard detect the CLI you are already logged into. Fetching starts paused until you finish the setup, so a blank profile never burns your quota.
Engines chain with automatic failover. Consumer-subscription terms do not spell out background services; read yours before making a subscription the primary engine.