The Vision: Free AI-Powered Job Search for Everyone

[santifer]

📌 Update (June 2026): this is the original vision post (April 2026) and its core promise stands — the local core is free forever. For where the project is heading now, including the optional shared layer, read Where career-ops is going.

The problem

Job searching is broken. Companies use AI to filter you out before a human ever sees your resume. Meanwhile, tools that help you fight back charge $20–140/month — to people who are, by definition, looking for income.

58% of professionals say finding a job has gotten significantly harder. Applications per role have doubled since 2022. And yet the "solutions" are paywalled.

That's not right.

What we're building

career-ops started as my personal system — I used it to evaluate 740+ job offers, generate 100+ tailored CVs, and land my current role. I open-sourced it because I believe finding a job should be free.

Today it's a powerful CLI tool for technical users. But the vision is bigger:

career-ops will become a free, local-first AI job search agent that anyone can run on their own computer — no API keys, no subscriptions, and everything stays on your machine by default.

Since April, this vision has grown a second leg: an optional, opt-in shared layer on top of the local core (details here). Whether you use it is your choice — your data and control stay yours either way. This was never about price; the core is free forever.

Roadmap

🟢 Now (what we're shipping)

• Community foundations — Code of Conduct, Governance, Contributor Ladder, Security policy
• Zero-token portal scanner (scan.mjs) — scans 73+ company APIs with zero LLM cost
• Language expansion — Japanese, Korean, Brazilian Portuguese modes live. German and French already shipping
• Security hardening across all scripts
• Issue templates and PR templates to make contributing easier

🔵 Next (what's coming)

• Free local model support — Ollama + LM Studio integration so you can run career-ops with Llama 3, Mistral, or Phi-4 on your own machine. Zero API cost, total privacy
• Claude Desktop integration — one-click install for semi-technical users
• Setup wizard — replace YAML config files with a simple guided flow. Paste your CV, pick your target roles, done
• CLI-agnostic engine — make the evaluation engine work with any AI backend via a simple env var

🟣 Later (the big vision)

• Desktop app — a native application (Tauri-based, not Electron) that anyone can download and use. No terminal required. Drag-and-drop your CV, paste a job URL, get a full evaluation + tailored PDF in 30 seconds
• Bundled local AI — the app ships with a model built-in. Download, open, use. That's it
• Every language, every market — localized modes for India, MENA, Southeast Asia, Latin America, Eastern Europe
• Personalized learning paths — when the system detects skill gaps, it helps you close them

Why this matters

Job searching is a moment of maximum vulnerability — and the people who need leverage most are the ones least able to pay for it. career-ops exists to raise the floor for everyone: a serious, open-source job search agent that runs on your own machine and is free forever.

Communities compound in a way no single product can — every contributor makes the tool better for the next person who arrives. That's the bet, and it's working.

How you can help

You don't need to write code to contribute. Check our Contributor Ladder — every role matters:

• Translate — add your language so people in your country can use it
• Test — try career-ops and report what's confusing or broken
• Build — help with local model integration, the desktop app, or the setup wizard
• Share — if career-ops helped you land an interview or a job, tell people about it

Read CONTRIBUTING.md to get started.

One more thing

I built this because I was tired of the job search grind. This is MIT-licensed, open source, and free — built by someone who's been exactly where you are.

Let's make job searching suck less — together.

— Santiago

[aa29zzjj]

Really cool idea — the agentic job search pipeline makes a lot of sense.

I managed to get Claude Code running, but a few things were a bit rough:

• Claude Code dependency (and subscription requirement) isn't obvious at first
• /career-ops isn't a standard CLI command, which was confusing initially
• Parsing job URLs seems fragile (LinkedIn search pages break the flow)

The concept is solid though — especially the CV tailoring + evaluation loop.

Would be great to see:

• Better onboarding / error handling
• Support for non-Claude backends

Thanks for open-sourcing this!

[santifer]

Thanks for your feedback! We are working on this; better onboarding and error handling are a must.

Career-ops actually supports non-Claude backend, which one are you missing?

[Deepakhacker-lab]

Great contribution

[santifer]

Thanks!

[KrAsH-CoD3]

Great work I must say.

[TheKelvinPerez]

This is a really cool project i'm going to give it a spin an see how it works

[shvetsov-as]

100% cool project and great work!

[itsmedhawal]

Hi @santifer :)
Incredible vision laid out here. The "free, local-first AI job search agent" direction is exactly where this needs to go.

I have been building a browser-based companion to career-ops-india (my India fork of career-ops) and wanted to share what I shipped this today. It directly mirrors what you are planning and might be useful to the community.

Career-Ops-India App v7.0 — itsmedhawal.github.io/career-ops-india-app

The first app in India to offer a hybrid Ghost Buster — local + cloud, mobile-first, privacy-first.

What is live today:

• ⚡ Fast Scan — instant local GLS via Gemma 2 2B running on-device via WebLLM + WebGPU. Zero API cost. Works offline after a one-time ~1.6 GB download.
• 🔍 Deep Eval — full A–G evaluation via Claude API when the role passes the GLS pre-check
• 🔒 Total privacy — JD text never leaves the device during Fast Scan. API key encrypted with PIN using AES-GCM.
• 📱 Mobile-first PWA — installable, works offline, optimised for one-thumb use on the go

The hybrid architecture:
Local AI acts as the First Responder — fast, free, offline GLS pre-check. If the role looks real, the user then spends API credits for the deep A–G evaluation and CV tailoring. This maps exactly to your "Zero token portal scanner + LLM evaluation" split.

On the Gemma 4 E2B roadmap:
Built with a model selector so the community can upgrade to Gemma 4 E2B the moment MLC compiles it for WebLLM — no code changes needed, just a model ID swap. Watching the MLC repo for that.

India-specific additions on top of career-ops:

• Ghost Likelihood Score (GLS) — original 0–100 detector across 9 India-calibrated signals (repost, Naukri source quality, fake remote, unnamed client, etc.)
• 7 India archetypes, CTC/LPA framework, GCC stage, Fake Remote detection
• Intern Mode — PPO probability, stipend benchmarks across 10 Indian cities
• Custom CV generator and Interview Prep Mode

The lineage is clear: career-ops → career-ops-india → Career-Ops-India App. Everything builds on the foundation you created. Thank you for open-sourcing it.

Would love to contribute the India module back to the main project when the time is right — especially GLS and the browser-based local AI architecture, which could serve as a reference for other regional implementations.

— Dhawal
career-ops-india · Career-Ops-India App

[itsmedhawal]

https://www.linkedin.com/posts/dhawalshrivastava_distributedintelligence-dtalks-careeropsindia-share-7448879949538545664-nDPN?utm_source=share&utm_medium=member_desktop&rcm=ACoAAATFzxkB0oXfvXocZiJc2pihpYhHVjEFims

[santifer]

Dhawal, nice work. The local AI pre-check with Gemma is a smart approach to the "free for real" problem. Running inference on-device means zero cost for the user, which is exactly the direction we laid out in the roadmap. The India-specific ghost detection with calibrated signals also adds real value for that market.

Good to see people building on career-ops and taking it to new regions. That's what open source is about.

Feel free to reach out if you want to chat more about what you're seeing in the Indian market.

[itsmedhawal]

@santifer sure thing :)

[itsmedhawal]

Hey @santifer, check this: https://coi-app.pages.dev
I would love to know your feedback.

Furthermore, Thank you for your invite as contributor. I believe I will soon be in a state to contribute INDIA specific changes upstream to career-ops :) from career-ops-india. Please let me know how we shall plan on this.

[DanielLin9406]

I completely support this idea. I recently had both of my accounts banned by Anthropic, even though I was only using Career-Ops. It's incredibly dangerous to risk vendor lock-in; a sudden ban can completely disrupt any workflow or progress that relies on a specific AI model.

[thegeekybeng]

Hi there, I have written the RFC for my suggestion to improve token consumption. And i had made a suggestion for a selection panel - users gets to choose if they would like to implement Career-Ops as per the original, or use a token-friendly approach what i denoted as zero-waste, to ensure better efficiency in consumption.

That being said, I think I can help with your project, as I had previously created and deployed The Perfect Fit, a career coach web tool that i shut down a couple of weeks ago when I noticed someone was using the capabilities of TPF to charge others for resume writing at S$400 per hour. I am in the midst of reviving TPF v11 - and perhaps we should connect.

[MyNameReallySux]

Hey Santiago 👋 — just wanted to say this vision really resonates. The local-first, "your data stays on your machine" approach is exactly what this space needs. Charging unemployed people $20–140/month for what is mostly prompting and pipeline glue never sat right with me. Happy to be contributing.

What I've been working on / planning to contribute:

• fix(merge-tracker): use token-union ratio in roleFuzzyMatch to stop cross-role dedup #793 (open PR) — fix for merge-tracker.mjs fuzzy dedup silently merging distinct roles at the same company (token-union/Jaccard ratio instead of min-side overlap). I have a follow-up ready that ports the same fix to dedup-tracker.mjs.
• feat(dashboard): hotkeys to view and (re)generate the CV PDF for the selected application #794 (open issue) — dashboard hotkeys to view (d) and regenerate (D) the CV PDF for the selected application, plus a small PDF index so the dashboard can resolve report → PDF path. Implementation is working on my fork; PR coming once it's had a bit more testing.
• npm scripts for the dashboard — build:dashboard / serve:dashboard so the Go TUI is launchable via npm run like every other tool in the project (and copy-pasteable on Windows). Issue write-up is drafted; will file it shortly.
• Smaller follow-ups from exercising the system on Windows: tracker link normalization in the dashboard (related to Tracker report links aren't clickable, written repo-root-relative instead of relative to the tracker file #760), and general Windows-path/PowerShell ergonomics as I hit them.

Thanks for open-sourcing this — looking forward to where it goes.

[santifer]

Hey @MyNameReallySux: overdue reply, and your comment deserved better than the wait (busy months around the releases). Thanks for saying it so plainly. "Charging unemployed people $20-140/month for prompting and pipeline glue" is exactly the thing this project exists to refuse, and that promise hasn't moved: the local core is free, MIT, forever.

Since you wrote this in June, the vision got more concrete in the two directions you care about:

• No vendor lock-in, for real. career-ops now runs on 9 AI CLI hosts through the open Agent Skill Standard (Claude Code, Codex, Gemini, OpenCode, Grok, Qwen and more), plus local-model paths: Ollama eval works today. @DanielLin9406's story above is why this is roadmap priority and not a nice-to-have: if one vendor bans you, you switch CLIs and keep your pipeline.
• Your data stays yours. The user/system boundary is now a written contract (DATA_CONTRACT.md) enforced by CI, and this week it grew fork-local declarations: files the updater is told never to touch because they're yours, not the project's.

And since you've shipped 10 merges here: if anything in the current direction itches, you know where the issues are. Good to see you around.

[mkriegl]

I am open to contribute for German translation. Furthermore I am actively looking for new opportunities, so due to active usage open for testing and reporting
I am no developer and still in the finding phase of using career-ops, but already excited what already has been developed.

[santifer]

Welcome, @mkriegl: testing and reporting from someone in the middle of a search is the most useful feedback this project gets (your doctor report became a fix within two days). For German, the concrete open pieces are #3406 (interview modes into German) and the German interview README in #3564: say which and it is yours.

[mkriegl]

I can take both. Tomorrow is one of the last warm days in Germany and going cycling, but due to rain from Wednesday I am able to step in.
From my employer I am released until end of the year, so I am using the agent actively to find something new, but in-between I am able to take over topics.
I put both on my list, if it is okay

[JOduMonT]

@santifer
as I mentionned in another discussion; I developped a plugin for...
before using career-ops I developped a job-digest for my city/country that literally scavenged more than 200 sites, deduplicated jobs based on profile (which I think you call archetype), then send me a daily 30 sec summary on what's new and what to apply.

so; instead of fighting with you on PR and slowing you're core development, and mostly because of...
I plug in my job-digest within your project.

But then; why the plugin is not yet already public?

first; on top of playwright I use browser and flaresolverr; which both have pros and cons but definitely take a route that user need to be aware of before (I means it is not abusing or bracking anything; but it definitely contouring some security).

In anycase;

1. my first question is about ethicaly speaking

since your project target end user (individuals) and is distribute MIT (PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND); do you think contouring security is ok and should I push my plugin to the community ?

2. as I mentionned my list is city, potentially country specifically targeted;

how in the future you see contributor to "specialize/localize" their own version of career-ops without breaking the opportunities to receive updates from the core project ?

[JOduMonT]