Job Agent

CLI tool that automates filling out online job application forms. Given a job posting URL and your profile/resume, it navigates the form, writes tailored cover letters and screening answers, and asks you only when it hits a genuine blocker (CAPTCHA, login wall, MFA).

Requirements

• Python 3.11+
• uv
• An API key for your chosen LLM provider (Google Gemini by default)

Setup

uv sync
uvx browser-use install
cp .env.example .env
# Edit .env — set the API key for your provider

Copy applicant/profile.yaml.example to applicant/profile.yaml and fill in your details. Add your resume as applicant/resume.md (Markdown preferred; .pdf also accepted).

To switch the LLM provider or model, edit _LLM_PROVIDER / _LLM_MODEL at the top of src/job_agent/main.py.

Best accuracy: browser-use's own models outperform generic frontier LLMs on web tasks.

• bu-ultra (cloud) — 78% accuracy; includes stealth browser, CAPTCHA solving, and optimized tool orchestration. Set _LLM_PROVIDER = "browseruse" and add BROWSER_USE_API_KEY to .env.
• ChatBrowserUse-2 (open-source) — 63.3% accuracy; self-hostable, faster and cheaper per task than Claude/GPT/Gemini.

Usage

# Fill a form (dry run — does not submit)
uv run job-agent apply https://jobs.example.com/apply/123 --dry-run

# Fill and submit
uv run job-agent apply https://jobs.example.com/apply/123

# Run browser headlessly
uv run job-agent apply <url> --headless

# View past applications
uv run job-agent log
uv run job-agent log --limit 20

# Validate your profile and resume files
uv run job-agent profile check

Always test with --dry-run before enabling submission on a new site.

Profile (applicant/profile.yaml)

The agent reads this file to fill every form field. The more complete it is, the fewer questions it needs to ask. Key sections:

personal:
  first_name: Jane
  last_name: Doe
  email: jane@example.com
  phone: "+1-555-555-5555"
  location: San Francisco, CA
  linkedin: https://linkedin.com/in/janedoe
  github: https://github.com/janedoe
  twitter: https://x.com/janedoe
  telegram: https://t.me/janedoe

target:
  roles:
    - Senior Software Engineer
    - Backend Engineer
  seniority: Senior
  industries:
    - fintech
    - developer tools
  preferred_locations:
    - San Francisco, CA
    - Remote
  open_to_relocation: false
  notice_period: "2 weeks"
  work_authorization:
    citizenship: US Citizen
    authorized: true
    sponsorship_needed: false

experience:
  years_total: 7
  current_title: Senior Software Engineer
  current_company: Acme Corp
  summary: >
    Backend-focused engineer with 7 years building distributed systems in Python and Go.

skills:
  languages:
    - Python
    - Go
    - TypeScript
  frameworks:
    - FastAPI
    - Django
    - React
  tools:
    - PostgreSQL
    - Redis
    - Kubernetes
    - Docker

education:
  - degree: B.S. Computer Science
    school: University of California, Berkeley
    year: 2018

compensation:
  desired_salary_usd: 180000
  open_to_equity: true

preferences:
  cover_letter_tone: professional   # professional | casual
  writing_style: natural-human      # natural-human | formal

answers:
  # Pre-baked answers used verbatim for common screening questions.
  why_this_company: >
    I'm drawn to the company's focus on developer experience and its track record
    of shipping pragmatic solutions to hard infrastructure problems.
  greatest_strength: >
    I break ambiguous problems into concrete pieces quickly.
  greatest_weakness: >
    I used to under-communicate blockers; I've since built a daily check-in habit.

For fields not in your profile, the agent generates a best-effort answer autonomously. It only pauses to ask you for hard blockers it cannot bypass.

Environment variables

Variable	Purpose
GOOGLE_API_KEY	Google Gemini (default provider)
OPENAI_API_KEY	OpenAI provider
ANTHROPIC_API_KEY	Anthropic provider
GROQ_API_KEY	Groq provider
BROWSER_USE_API_KEY	browseruse cloud provider
BROWSER_HEADLESS	true/false — headless mode
LOG_DB_PATH	SQLite log path (default: ./applications.db)

Related projects

Other open-source job-application tools, for comparison:

Project	Stars	Approach	Notes
AIHawk	~30k	LLM-driven, tailored answers	Most popular; archived/unmaintained
Auto_job_applier_linkedIn	~2.4k	Selenium, LinkedIn Easy Apply	Best actively-maintained option
ApplyPilot	~1.1k	Agentic, "any site, any form"	Closest in ambition to this project
EasyApplyJobsBot	~0.8k	Scripted, multi-board	LinkedIn, Glassdoor, Greenhouse, etc.
career-ops	—	Claude Code-based	A–F scoring, ATS PDF gen, scans Greenhouse/Ashby/Lever

Most popular tools are scripted Selenium bots targeting LinkedIn Easy Apply (which bans automation). This project is distinct in being browser-use + LLM-agent-driven, working on any form autonomously.

Development

uv run mypy src/          # type-check (must pass clean)
uv run ruff check --fix . # lint
uv run ruff format .      # format
uv run pytest             # tests