currobot

Local-first job application automation for the Spanish job market. Runs as a desktop app on macOS and Windows.

Scrapes 14+ job boards, adapts your CV for each role using a local AI model, and submits applications — but only after you approve each one. Nothing leaves your machine.

---

What it does

1. Scrapes InfoJobs, Indeed ES, Mercadona, Lidl, Amazon ES, Greenhouse, Lever, TeamTailor, Personio, Workday, Manfred, Tecnoempleo, JobToday, and custom career pages on a schedule
2. Filters jobs that explicitly disqualify for the student→work permit canje (temporal contracts, part-time, salary below SMI €15,876/year)
3. Adapts your CV using a local Ollama model — reframes your experience for each job profile (cashier, stocker, logistics, frontend dev, fullstack dev) without inventing anything
4. Fills the application form using browser automation with human-like behaviour
5. Asks you to approve — you see the filled form + adapted CV before anything is submitted. No submission happens without your click.

---

Install

Download the latest release for your platform from the Releases page:

• macOS — .dmg → drag currobot to Applications
• Windows — .msi → run the installer

macOS only — "cannot be opened" warning: macOS may block currobot because it isn't signed with an Apple Developer certificate. To open it anyway, right-click (or Control-click) currobot.app and choose Open, then click Open in the dialog. You only need to do this once.

Alternatively, run this in Terminal after installing:

xattr -d com.apple.quarantine /Applications/currobot.app

Before launching, install Ollama (the local AI runtime): Download from ollama.ai and run it. currobot uses it to adapt your CV locally — no data leaves your device.

---

First launch

The first time you open currobot, a setup wizard guides you through:

1. System check (Ollama running, enough disk space)
2. RAM check + AI model recommendation
3. Download the Ollama model (one-time, ~4 GB)
4. Upload your master CV as a PDF
5. Accept the terms
6. Choose whether to start currobot on login

After that you're in the dashboard. The Python backend and scrapers run automatically in the background.

---

Daily use

currobot lives in the system tray. Close the window and it keeps running. Click the tray icon to reopen.

Page	What it does
Dashboard	System health, scraper status, application funnel
Jobs	Browse scraped jobs, filter by site or profile
Applications	Kanban board — from scraped to applied
Review	Approve pending applications before submission (30-min window)
CV	Manage your master CV
Settings	Ollama model, retention, company sources, start-on-login toggle

Tray menu:

• Open currobot — bring the window to front
• Start on Login — toggle autolaunch (checkmark = enabled)
• Uninstall currobot… — disables autolaunch then quits; then delete the app manually

---

Uninstall

macOS — one-liner (removes app, all data, and Keychain entries):

osascript -e 'tell application "currobot" to quit' 2>/dev/null; sleep 1; \
  rm -rf /Applications/currobot.app \
         "$HOME/Library/Application Support/com.currobot.app" \
         "$HOME/Library/Logs/com.currobot.app"; \
  security delete-generic-password -s currobot 2>/dev/null; \
  launchctl remove com.currobot.app 2>/dev/null; \
  echo "currobot removed"

Windows — run in PowerShell as Administrator:

Stop-Process -Name currobot -Force -ErrorAction SilentlyContinue
Start-Sleep 1
Get-WmiObject Win32_Product | Where-Object { $_.Name -eq "currobot" } | ForEach-Object { $_.Uninstall() }
Remove-Item "$env:APPDATA\com.currobot.app" -Recurse -Force -ErrorAction SilentlyContinue
cmdkey /delete:currobot 2>$null
Write-Host "currobot removed"

Manual steps (if you prefer):

1. Right-click the tray icon → Uninstall currobot… — disables autolaunch and quits the app
2. Remove the app:
   • macOS: move currobot.app from Applications to Trash
   • Windows: Settings → Add or Remove Programs → currobot → Uninstall
3. Remove app data (optional):
   • macOS: ~/Library/Application Support/com.currobot.app/
   • Windows: %APPDATA%\com.currobot.app\
4. Remove Keychain entries (optional):
   • macOS: Keychain Access → search "currobot" → delete
   • Windows: Credential Manager → remove "currobot" entries

---

What the filters do

currobot automatically skips jobs that explicitly disqualify for the Spanish student→work permit canje (Reglamento de Extranjería 2025):

Disqualifier	Example
Temporal contract	"contrato temporal", "fijo discontinuo", "interinidad", "eventual", "por obra"
Part-time	"media jornada", "tiempo parcial", "20 horas semanales"
Salary below SMI	Any stated salary below €15,876/year or €1,134/month

If no contract type or salary is mentioned, the job passes through — the filter only acts when something is explicitly disqualifying. Update SMI_MONTHLY_GROSS and SMI_ANNUAL_GROSS in backend/scrapers/visa_filter.py each January when the new SMI is published.

---

For developers

Tech stack

Layer	What
Desktop shell	Tauri v2 (Rust) — tray, window, autolaunch, notifications
Backend	Python 3.11+, FastAPI, SQLAlchemy 2 (async), aiosqlite
Scraping	Patchright (stealth Playwright), BeautifulSoup4, httpx
AI	Ollama (local LLM), custom 4-step CV adaptation pipeline
PDF	pdfplumber (parse), reportlab + weasyprint (generate)
Scheduling	APScheduler
Frontend	Next.js 16 (static export), React 19, Tailwind CSS 4, Radix UI
Real-time	Server-Sent Events (SSE)
Credentials	keyring (macOS Keychain / Windows Credential Manager / Linux Secret Service)
Tests	pytest

Project structure

jobbot/
├── backend/
│   ├── main.py                  # FastAPI app, all REST + SSE endpoints
│   ├── config.py                # Central config, rate limits, data paths
│   ├── database/
│   │   ├── models.py            # SQLAlchemy ORM models + status enums
│   │   ├── crud.py              # All DB writes (single-writer pattern)
│   │   ├── session.py           # AsyncSessionLocal factory
│   │   └── alembic/             # Migrations
│   ├── scrapers/
│   │   ├── base.py              # BaseScraper — all scrapers inherit this
│   │   ├── visa_filter.py       # Canje eligibility filter (contract/salary/hours)
│   │   ├── scheduler.py         # APScheduler job registration
│   │   ├── browser_pool.py      # Shared Playwright contexts + cookie persistence
│   │   └── *.py                 # 14 site-specific scrapers
│   ├── ai/
│   │   ├── cv_adapter.py        # 4-step CV adaptation pipeline
│   │   ├── ollama_client.py     # HTTP client to Ollama
│   │   ├── prompts.py           # All LLM prompt templates (versioned)
│   │   ├── quality_check.py     # Score adapted CV against job description
│   │   └── validator.py         # Anti-fabrication checks
│   ├── application/
│   │   ├── form_detector.py     # Detect and classify form fields
│   │   ├── form_filler.py       # Fill fields with semantic mapping (ES/EN)
│   │   ├── confirm_detector.py  # Detect confirmation screens post-submit
│   │   └── human_loop.py        # Prepare review, auth flow, 30-min timeout
│   ├── notifications/
│   │   └── notifier.py          # In-memory notification queue (SSE → Tauri)
│   └── documents/
│       ├── cv_parser.py         # Parse PDF CV → canonical JSON
│       └── cv_generator.py      # Canonical JSON → adapted PDF
├── frontend/
│   ├── app/                     # Next.js App Router pages
│   ├── components/              # Sidebar, CommandPalette, UI primitives
│   ├── lib/                     # API client, types, utilities
│   └── src-tauri/               # Tauri desktop shell (Rust)
│       ├── src/lib.rs           # Tray, window, sidecar, autolaunch logic
│       ├── tauri.conf.json      # App config, bundle settings
│       └── capabilities/        # Tauri v2 permission definitions
├── tests/
│   └── test_visa_filter.py      # 88 pytest cases for the visa filter
├── backend.spec                 # PyInstaller spec (bundles backend into binary)
├── Makefile                     # Build pipeline
├── requirements.txt
└── .env.example                 # Config reference

Running in development

# Option 1 — backend + Next.js dev server (no native window)
make dev

# Option 2 — full Tauri desktop window
make tauri-dev

make dev requires Python 3.11+ and Node 20+ installed locally. Ollama must be running.

Building a distributable

make build

This runs PyInstaller to bundle the Python backend, then tauri build to produce the platform installer (.dmg on macOS, .msi on Windows).

Running tests

python3 -m pytest tests/ -v

Currently tested: visa_filter.py (88 cases covering contract type, part-time, salary parsing, edge cases, and combined scenarios).

Adding a scraper

1. Create backend/scrapers/yoursite.py — inherit from BaseScraper, implement scrape() -> list[dict]
2. Each job dict must include: external_id, url, title, company. Optional: location, description, salary_raw, contract_type, cv_profile
3. Register it in backend/scrapers/scheduler.py → _get_scraper_map()
4. Add its rate limit to RATE_LIMITS in backend/config.py

The base class handles: deduplication, visa filter, rate limiting, consecutive-zero guard, scraper run logging, and SSE broadcast.

Adding a CV profile

1. Add the enum value to CVProfile in backend/database/models.py
2. Add the profile config to PROFILE_REFRAME in backend/ai/cv_adapter.py
3. Run a database migration: alembic revision --autogenerate -m "add profile" + alembic upgrade head

Configuration

Copy .env.example to .env and edit as needed:

OLLAMA_HOST=http://localhost:11434   # change if Ollama runs on another machine
OLLAMA_MODEL=                        # leave empty for auto-selection by RAM
SCRAPER_DEFAULT_DELAY_MIN=3.0
SCRAPER_DEFAULT_DELAY_MAX=8.0
JOBS_RETENTION_DAYS=90
APPLICATIONS_RETENTION_DAYS=365

Credentials (job site logins) are stored in the OS keychain via keyring — never in .env.

Database

SQLite at data/jobs.db. Migrations with Alembic:

alembic upgrade head
alembic revision --autogenerate -m "description"

---

Privacy

• All CV data, job data, and application history stays on your machine in data/
• The AI runs locally via Ollama — no data is sent to any external service unless you point OLLAMA_HOST at a remote server
• Browser sessions and cookies are stored locally in data/browser_profiles/
• The data/ directory is excluded from git

---

License

MIT