Self-hosted, multi-account job-hunt automation.
Scrape career pages and aggregators, score jobs against your résumés with an LLM, tailor résumés and cover letters, auto-fill applications from your persona, get Telegram alerts, track every application. One install serves one person or a household: up to five accounts, each with its own jobs, résumés, settings and optional PIN, sharing one scrape.
Updating from 2.x
Follow steps here manually or ask AI to: Upgrading from 2.x
Five themes, each in light and dark: Paper (above), Green Paper, Stone, V1 Style, Windows 98.
┌─────────────────────────────────────────────────────────────────────────────┐
│ JOB DISCOVERY │
│ │
│ Career Pages │ Aggregators │ Chrome Extension │
│ │ │ │
│ Any site via │ JobSpy: LinkedIn, │ Passive LinkedIn │
│ Playwright │ Indeed, ZipRecruiter, │ capture while │
│ │ Google Jobs │ browsing │
│ 12 ATS endpoints: │ │ │
│ Workday, Greenhouse │ LinkedIn Personal │ Auto-fill based on │
│ Lever, Ashby, │ collections │ your persona input │
│ Oracle, Phenom, │ Apify (paid boards) │ and question bank │
│ TalentBrew, Rippling │ Jobright.ai │ │
│ SmartRecruiters, │ Levels.fyi │ Save any job from │
│ Amazon, Meta, Google │ freehire.me │ any page │
│ │ CaribbeanJobs.com │ │
│ │ TrulyRemoteWork.com │ │
│ │ │ │
└─────────────────────────┴───────────┬─────────────┴─────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────────┐
│ PROCESSING │
│ │
│ Dedup ────── URL-hash dedup, tracking params stripped │
│ Filters ──── Title / company include & exclude, body exclusion phrases │
│ H-1B ─────── Company LCA data from MyVisaJobs (cached) │
│ Salary ───── Extracted from posting, H-1B data, description │
│ │
└─────────────────────────────────────┬───────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────────┐
│ JOB FEED │
│ │
│ Review ───── Dynamic filters, sorting, detail panel │
│ Decide ───── Save promising jobs, skip the rest, score with AI │
│ │
└─────────────────────────────────────┬───────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────────┐
│ AI RESUME SCORING │
│ │
│ Providers ── Claude API/CLI, Codex CLI, Antigravity CLI, OpenCode Go, │
│ OpenAI, OpenRouter, Ollama, LM Studio │
│ Depths ───── Light (scores only) or Full (report + keyword analysis) │
│ Multi ────── Score against multiple resumes, compare fit per role │
│ │
└─────────────────────────────────────┬───────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────────┐
│ RESUME + COVER LETTER │
│ │
│ Templates ── 8 resume + 7 cover-letter, auto-discovered (add your own!) │
│ AI Tailor ── Rewrites resume bullets/keywords from the scoring report │
│ AI Letter ── Job-specific cover letters from resume + JD, voice presets │
│ Export ───── Live-preview PDF via Playwright with export │
│ │
└─────────────────────────────────────┬───────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────────────────────┐
│ TRACK │
│ │
│ Autofill ─── ATS forms + AI answers to free-text questions (Persona) │
│ Apps ─────── Stages, interviews, notes, prep handover, funnel and Sankey │
│ Tracer ───── Unique links per resume/letter, tracks who opened them │
│ Gmail ────── Auto-detects responses, updates application status │
│ Telegram ─── Job alerts, daily digest, scrape health notifications │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
| Feature | Description |
|---|---|
| Discovery | Career pages (Playwright + 12 ATS handlers), JobSpy (LinkedIn, Indeed, ZipRecruiter, Google), Apify (Indeed, LinkedIn, Glassdoor, Greenhouse/Lever/Ashby/Workday search; not on the Apify Creator plan), LinkedIn collections, Levels.fyi, Jobright.ai, freehire.me, CaribbeanJobs.com, TrulyRemoteWork.com, the Chrome extension |
| Scoring | Claude, OpenAI, OpenRouter, Ollama, LM Studio, Claude Code, Codex CLI, Antigravity CLI or OpenCode Go; live model search, a model per feature; Light (score) or Full (report, keyword coverage, requirement mapping) against every base résumé; prompt caching on Anthropic |
| Résumés | Structured base résumés, tailoring per job with a review step, 8 PDF templates (drop in your own) and a Word (.docx) export, an opt-in page footer (name, job title, company) on tailored copies, tracked links that record opens |
| Cover letters | Generated from the paired résumé and persona; voice and length presets, 7 templates, PDF |
| Dedup | URL identity hash (tracking params stripped) plus company + title hash across sources |
| Jobs | Filters, sorting, keyboard shortcuts (?), full report, collapsible analysis pane, bulk actions with undo, posting preview |
| Applications | Stages, interviews, notes, status history (funnel and Sankey in Stats), a prep handover for any AI chat |
| Persona | Contact, work authorization, compensation, preferences, résumé content, Q&A bank; import from a résumé or PDF |
| Extension | Passive LinkedIn capture, save any job from any page, ATS form autofill and AI answers to free-text questions from your Persona |
| Accounts | Owner + up to four members, each with their own jobs, applications, résumés, letters, Persona, searches and companies; optional 4-digit PINs; settings inherited or own per section; scraping shared (one fetch per page, filtered and scored per account); with one account none of it shows |
| Gmail | Polls replies, classifies them, updates the application; each account connects its own mailbox |
| Telegram | New-job alerts, daily digest, scrape health, inline actions; per account (household bot or its own) |
| H-1B | Company filing lookups, description exclusion phrases |
| Scheduling | Intervals and crons for scraping, email, backups, cleanup, auto-reject; every cron field explains itself and offers presets |
| Themes | Paper, Green Paper, Stone, V1 Style, Windows 98 × light / dark / system |
The posting preview is an
iframe; sites that refuse framing show blank unless the extension (which strips the frame-blocking headers) is installed. "Open" always works, and applied jobs keep a cached snapshot.
1. Run it
git clone https://github.com/vesaias/JobNavigator.git
cd JobNavigator
cp .env.example .env
docker compose pull && docker compose up -d # prebuilt images from GHCROpen http://localhost. To build from source instead (or on ARM), use docker compose up --build -d; every release is also published as ghcr.io/vesaias/jobnavigator-backend and -frontend, pinnable with JN_IMAGE_TAG=3.0 in .env. On first run sign in with a blank key, then set one in Settings › System › Advanced.
2. Connect a model
Any API key goes straight into Settings › AI. A subscription needs one login inside the container first (OpenCode Go needs none):
| Subscription | Once after the containers start |
|---|---|
| Claude Pro / Max | docker compose exec backend claude setup-token, put the printed token in .env as CLAUDE_CODE_OAUTH_TOKEN, then docker compose up -d backend |
| ChatGPT | docker compose exec backend codex login --device-auth (check with codex login status; the login lives in the codex_auth volume) |
| Google Antigravity | docker compose exec -it backend agy — the container has no browser, so the CLI prints a URL and asks for the code it shows (the login lives in the antigravity_auth volume) |
| OpenCode Go | nothing: paste the Go API key under Settings › AI |
No pay-per-token key needed for any of them. A plan limit fails over to the fallback provider without retrying. With several accounts, docker compose exec backend python -m backend.scripts.cli_login <codex|agy|claude> "<account>" prints a member's own login command.
3. First steps
- Settings › AI — provider and model
- Résumés — create a base résumé or import a PDF
- Persona — import from that résumé, fill the Q&A bank
- Companies and Searches — add a few, run them
The full walkthrough, screen by screen, is in docs/GETTING-STARTED.md.
-
Unblocks the posting preview: frame-blocking headers are lifted only for frames the dashboard itself opens (scoped to your JobNavigator host); every other page keeps its own headers.
-
Captures job ids while you browse
linkedin.com/jobs/collections/*and imports them with full details. -
Fills ATS forms and drafts answers to free-text questions from your Persona and Q&A bank (toggle in the popup; model and prompt in Settings › AI).
-
With two or more accounts, Saves to in the popup picks the account that captures and autofill use.
Install: chrome://extensions › Developer mode › Load unpacked › extension/. The LinkedIn import signs in through Settings › Integrations › LinkedIn (a separate mock account is recommended).
Server (.env) |
Per account (Settings) | |
|---|---|---|
| Telegram | TELEGRAM_BOT_TOKEN (the household bot) |
Notifications › Link Telegram, or its own bot |
| Gmail | GMAIL_CLIENT_ID / GMAIL_CLIENT_SECRET; add <dashboard>/api/gmail/callback as a redirect URI |
Email classification › Gmail › Connect Gmail |
| Layer | Technology |
|---|---|
| Backend | Python 3.12, FastAPI, SQLAlchemy, APScheduler, Playwright |
| Frontend | React 18, Vite, Recharts, a token-based design system (DESIGN-SYSTEM.md) |
| Database | PostgreSQL 16 |
| Infrastructure | Docker Compose, Caddy, nginx |
| AI | Anthropic SDK, OpenAI SDK, Ollama, LM Studio, Claude Code CLI, Codex CLI, Antigravity CLI, OpenCode Go endpoint |
| Extension | Chrome Manifest V3 |
Scheduled pg_dumps land in backups/ (five kept). They contain the settings table, API keys included: treat them like .env. The folder is git-ignored.
Self-hosted. Your résumés, jobs and credentials stay on your machine; the only outside party is the AI provider you configure.
Personal use. Not affiliated with any job platform; some scrapers are off by default and you are responsible for the terms of the sites you use. See LEGAL_DISCLAIMER.md.