ai-job-application-agent

⚠️ ApplyPilot is the original open-source project, created by Pickle-Pixel and first published on GitHub on February 17, 2026. We are not affiliated with applypilot.app, useapplypilot.com, or any other product using the "ApplyPilot" name. These sites are not associated with this project and may misrepresent what they offer. If you're looking for the autonomous, open-source job application agent — you're in the right place.

ApplyPilot

Applied to 1,000 jobs in 2 days. Fully autonomous. Open source.

https://github.com/user-attachments/assets/7ee3417f-43d4-4245-9952-35df1e77f2df

---

What It Does

ApplyPilot is a 6-stage autonomous job application pipeline. It discovers jobs across 5+ boards, scores them against your resume with AI, tailors your resume per job, writes cover letters, and submits applications for you. It navigates forms, uploads documents, answers screening questions, all hands-free.

Three commands. That's it.

pip install applypilot
pip install --no-deps python-jobspy && pip install pydantic tls-client requests markdownify regex
applypilot init          # one-time setup: resume, profile, preferences, API keys
applypilot doctor        # verify your setup — shows what's installed and what's missing
applypilot run           # discover > enrich > score > tailor > cover letters
applypilot run -w 4      # same but parallel (4 threads for discovery/enrichment)
applypilot apply         # autonomous browser-driven submission
applypilot apply -w 3    # parallel apply (3 Chrome instances)
applypilot apply --dry-run  # fill forms without submitting

Why two install commands? python-jobspy pins an exact numpy version in its metadata that conflicts with pip's resolver, but works fine at runtime with any modern numpy. The --no-deps flag bypasses the resolver; the second command installs jobspy's actual runtime dependencies. Everything except python-jobspy installs normally.

---

Two Paths

Full Pipeline (recommended)

Requires: Python 3.11+, Node.js (for npx), Gemini API key (free), Claude Code CLI, Chrome

Runs all 6 stages, from job discovery to autonomous application submission. This is the full power of ApplyPilot.

Discovery + Tailoring Only

Requires: Python 3.11+, Gemini API key (free)

Runs stages 1-5: discovers jobs, scores them, tailors your resume, generates cover letters. You submit applications manually with the AI-prepared materials.

---

The Pipeline

Stage	What Happens
1. Discover	Scrapes 5 job boards (Indeed, LinkedIn, Glassdoor, ZipRecruiter, Google Jobs) + 48 Workday employer portals + 30 direct career sites
2. Enrich	Fetches full job descriptions via JSON-LD, CSS selectors, or AI-powered extraction
3. Score	AI rates every job 1-10 based on your resume and preferences. Only high-fit jobs proceed
4. Tailor	AI rewrites your resume per job: reorganizes, emphasizes relevant experience, adds keywords. Never fabricates
5. Cover Letter	AI generates a targeted cover letter per job
6. Auto-Apply	Claude Code navigates application forms, fills fields, uploads documents, answers questions, and submits

Each stage is independent. Run them all or pick what you need.

---

ApplyPilot vs The Alternatives

Feature	ApplyPilot	AIHawk	Manual
Job discovery	5 boards + Workday + direct sites	LinkedIn only	One board at a time
AI scoring	1-10 fit score per job	Basic filtering	Your gut feeling
Resume tailoring	Per-job AI rewrite	Template-based	Hours per application
Auto-apply	Full form navigation + submission	LinkedIn Easy Apply only	Click, type, repeat
Supported sites	Indeed, LinkedIn, Glassdoor, ZipRecruiter, Google Jobs, 46 Workday portals, 28 direct sites	LinkedIn	Whatever you open
License	AGPL-3.0	MIT	N/A

---

Requirements

Component	Required For	Details
Python 3.11+	Everything	Core runtime
Node.js 18+	Auto-apply	Needed for npx to run Playwright MCP server
Gemini API key	Scoring, tailoring, cover letters	Free tier (15 RPM / 1M tokens/day) is enough
Chromium-based browser	Auto-apply	Chrome, Arc, Brave, Chromium, and Edge are supported
Claude Code CLI	Auto-apply with APPLYPILOT_APPLY_BACKEND=claude	Install from claude.ai/code
Command backend	Auto-apply with APPLYPILOT_APPLY_BACKEND=command	Set APPLYPILOT_AGENT_COMMAND to your local or third-party apply agent

Gemini API key is free. Get one at aistudio.google.com. OpenAI and local models (Ollama/llama.cpp) are also supported.

Optional

Component	What It Does
CapSolver API key	Solves CAPTCHAs during auto-apply (hCaptcha, reCAPTCHA, Turnstile, FunCaptcha). Without it, CAPTCHA-blocked applications just fail gracefully

Note: python-jobspy is installed separately with --no-deps because it pins an exact numpy version in its metadata that conflicts with pip's resolver. It works fine with modern numpy at runtime.

---

Configuration

All generated by applypilot init:

profile.json

Your personal data in one structured file: contact info, work authorization, compensation, experience, skills, resume facts (preserved during tailoring), and EEO defaults. Powers scoring, tailoring, and form auto-fill.

searches.yaml

Job search queries, target titles, locations, boards. Run multiple searches with different parameters.

.env

API keys and runtime config: GEMINI_API_KEY, LLM_MODEL, CAPSOLVER_API_KEY (optional). Optional stage-specific model overrides are also supported: SCORING_LLM_MODEL, TAILOR_LLM_MODEL, COVER_LLM_MODEL.

Auto-apply backend selection also lives here:

• APPLYPILOT_APPLY_BACKEND=claude uses Claude Code (default)
• APPLYPILOT_APPLY_BACKEND=command runs APPLYPILOT_AGENT_COMMAND
• APPLYPILOT_AGENT_COMMAND supports {model}, {mcp_config}, {worker_dir}, {port}, {worker_id}
• scripts/local_apply_agent.py is a local browser-driving agent for Workday and Greenhouse flows
• APPLYPILOT_ACCOUNT_PASSWORD can be scoped with APPLYPILOT_ACCOUNT_PASSWORD_HOSTS
• APPLYPILOT_ACCOUNT_PASSWORDS_FILE supports per-host passwords without putting them in git
• APPLYPILOT_IMAP_* enables mailbox polling for password-reset and verification emails
• ~/.applypilot/question_memory/ stores per-company ATS question memory learned from successful and diagnostic runs

Package configs (shipped with ApplyPilot)

• config/employers.yaml - Workday employer registry (48 preconfigured)
• config/sites.yaml - Direct career sites (30+), blocked sites, base URLs, manual ATS domains
• config/searches.example.yaml - Example search configuration

---

How Stages Work

Discover

Queries Indeed, LinkedIn, Glassdoor, ZipRecruiter, Google Jobs via JobSpy. Scrapes 48 Workday employer portals (configurable in employers.yaml). Hits 30 direct career sites with custom extractors. Deduplicates by URL.

Enrich

Visits each job URL and extracts the full description. 3-tier cascade: JSON-LD structured data, then CSS selector patterns, then AI-powered extraction for unknown layouts.

Score

AI scores every job 1-10 against your profile. 9-10 = strong match, 7-8 = good, 5-6 = moderate, 1-4 = skip. Only jobs above your threshold proceed to tailoring.

Tailor

Generates a custom resume per job: reorders experience, emphasizes relevant skills, incorporates keywords from the job description. Your resume_facts (companies, projects, metrics) are preserved exactly. The AI reorganizes but never fabricates.

Cover Letter

Writes a targeted cover letter per job referencing the specific company, role, and how your experience maps to their requirements.

Auto-Apply

The Claude backend launches Chrome plus the Playwright MCP stack. The local command backend can drive Workday and Greenhouse flows directly with Playwright, fill personal information and work history, upload the tailored resume and cover letter, handle account bootstrap, use mailbox verification flows when needed, and submit when the form flow is stable. A live dashboard shows progress in real-time, and the HTML dashboard now refreshes automatically after prep and apply runs.

The Playwright MCP server is configured automatically at runtime per worker. No manual MCP setup needed.

# Utility modes (no Chrome/Claude needed)
applypilot apply --mark-applied URL    # manually mark a job as applied
applypilot apply --mark-failed URL     # manually mark a job as failed
applypilot apply --reset-failed        # reset all failed jobs for retry
applypilot apply --gen --url URL       # generate prompt file for manual debugging

---

CLI Reference

applypilot init                         # First-time setup wizard
applypilot doctor                       # Verify setup, diagnose missing requirements
applypilot run [stages...]              # Run pipeline stages (or 'all')
applypilot run --workers 4              # Parallel discovery/enrichment
applypilot run --stream                 # Concurrent stages (streaming mode)
applypilot run --min-score 8            # Override score threshold
applypilot run --dry-run                # Preview without executing
applypilot run --validation lenient     # Relax validation (recommended for Gemini free tier)
applypilot run --validation strict      # Strictest validation (retries on any banned word)
applypilot apply                        # Launch auto-apply
applypilot apply --workers 3            # Parallel browser workers
applypilot apply --dry-run              # Fill forms without submitting
applypilot apply --continuous           # Run forever, polling for new jobs
applypilot apply --headless             # Headless browser mode
applypilot apply --url URL              # Apply to a specific job
applypilot status                       # Pipeline statistics
applypilot report --section all         # Structured runtime + pipeline report
applypilot report --section ready --format markdown
applypilot dashboard                    # Open HTML results dashboard

The HTML dashboard is written to ~/.applypilot/dashboard.html. It is regenerated automatically by the prep/apply runners, and you can still force a refresh manually with:

cd /Users/yashm/Documents/ai-job-application-agent
PYTHONPATH=src python3 -m applypilot.cli dashboard
open ~/.applypilot/dashboard.html

Run Without Codex

Manual daily run from this repo:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/run_daily.sh

The runner does:

python3 scripts/daily_concurrent.py
applypilot status

Default knobs:

• APPLYPILOT_DAILY_MIN_SCORE=8
• APPLYPILOT_DAILY_SCORE_LIMIT=90
• APPLYPILOT_DAILY_TAILOR_LIMIT=35
• APPLYPILOT_DAILY_COVER_LIMIT=35
• APPLYPILOT_DAILY_TARGET_SUBMISSIONS=25
• APPLYPILOT_DAILY_APPLY_BATCH=5
• APPLYPILOT_DAILY_MAX_CYCLES=6
• APPLYPILOT_DAILY_VALIDATION=lenient
• APPLYPILOT_DAILY_WORKERS=1
• APPLYPILOT_APPLY_RETRY_COOLDOWN_HOURS=18
• APPLYPILOT_SUPPORTED_AUTOAPPLY_PATTERNS=myworkdayjobs.com,greenhouse.io,...
• APPLYPILOT_PREP_PREFERRED_SITES=greenhouse.io,BMO,RBC,Cisco,...
• APPLYPILOT_PREP_DEPRIORITIZE_SITES=Netflix,Adobe,Thomson Reuters,Moderna

Why this is better:

• discovery and enrichment still run daily for fresh jobs
• scoring, tailoring, and cover-letter generation are budgeted instead of trying to clear the whole backlog
• prep work and apply work now run in parallel loops during the same daily session
• the controller loops toward completed submissions rather than firing one big apply burst
• failed jobs cool down before retry, so fresh ready jobs get priority
• Greenhouse and other easier supported ATS targets can be preferred ahead of sticky tenants via env hints
• the run is quieter and more likely to reach the apply stage every day on a smaller local machine

Override them for a single run:

APPLYPILOT_DAILY_APPLY_LIMIT=5 ./scripts/run_daily.sh

Check whether the daily pipeline is currently running:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/is_running.sh

Get a Terminal summary of sourced jobs, daily pipeline activity, ready-to-apply jobs, and recent apply outcomes:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/check_applications.sh

Open the numbered local control menu:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/job_agent_menu.sh

Emit structured reports for automation, OpenClaw, or shell scripting:

cd /Users/yashm/Documents/ai-job-application-agent
PYTHONPATH=src python3 -m applypilot.cli report --section all --format json
PYTHONPATH=src python3 -m applypilot.cli report --section runtime --format table
PYTHONPATH=src python3 -m applypilot.cli report --section ready --format markdown

Run the system in true always-on mode:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/reload_always_on.sh

That stops the noon-only batch launcher, starts a detached always-on supervisor immediately, and keeps restarting the concurrent session after each batch completes.

Install login-time auto-start for the always-on supervisor:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/install_always_on.sh

Stop the always-on supervisor:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/stop_always_on.sh

Inspect learned ATS/company question memory:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/show_question_memory.sh --list
./scripts/show_question_memory.sh --ats greenhouse --list
./scripts/show_question_memory.sh "Grafana Labs"
./scripts/show_question_memory.sh "Thomson Reuters" --ats workday

Schedule Daily On macOS

Install a launchd job that runs every day at 12:00 PM local time:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/install_launchd.sh 12:00

Useful commands:

launchctl kickstart -k gui/$(id -u)/com.ai-job-application-agent.daily
launchctl print gui/$(id -u)/com.ai-job-application-agent.daily
launchctl bootout gui/$(id -u) ~/Library/LaunchAgents/com.ai-job-application-agent.daily.plist

After editing ~/.applypilot/.env, profile.json, or searches.yaml, the safest way to refresh the scheduled setup is:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/reload_daily.sh

If you want to reload the config and start a fresh run immediately:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/reload_daily.sh --run-now

Logs land in:

~/.applypilot/logs/

At the start of each scheduled run, the previous launcher logs are archived to:

~/.applypilot/logs/archive/

How launchd works here:

• launchd does not understand your job pipeline itself.
• It schedules a tiny wrapper under ~/.applypilot/bin/ that opens Terminal and runs scripts/run_daily.sh.
• That script runs scripts/daily_concurrent.py, which does the orchestration.
• This Terminal-backed launch is intentional. On macOS, background agents often cannot read repos under ~/Documents, while Terminal can because it runs with your normal interactive permissions.
• The daily supervisor now uses two loops at the same time:
• prep loop: discover/enrich/score/tailor/cover/pdf
• apply loop: applies ready jobs in batches
• So yes, now it can prepare later jobs while applying already-ready jobs in parallel during the same session.
• With APPLYPILOT_DAILY_WORKERS=1, apply itself is single-browser-worker. If you raise workers later, apply can parallelize across multiple ready jobs.

OpenClaw Wrappers

If you want OpenClaw on top, use it as a thin command runner over this repo.

Wrapper commands:

cd /Users/yashm/Documents/ai-job-application-agent
zsh ./scripts/openclaw_status.sh
zsh ./scripts/openclaw_activity.sh
zsh ./scripts/openclaw_jobs.sh ready
zsh ./scripts/openclaw_control.sh reload-always-on

What these do:

• openclaw_status.sh: prints runtime, overview, queue, failures, config, and history in Markdown
• openclaw_activity.sh: prints daily activity plus recent session events in Markdown
• openclaw_jobs.sh: prints ready jobs, recent apply activity, or failure lists in Markdown
• openclaw_control.sh: runs runtime controls and returns compact Markdown/text

Recommended OpenClaw usage:

• Register this repo's skills into OpenClaw:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/install_openclaw_integration.sh

• Then start a new OpenClaw session and use:
  • openclaw dashboard
  • /skill job-agent-help
  • /skill job-agent-status
  • /skill job-agent-activity
  • /skill job-agent-jobs ready
  • /skill job-agent-control reload-always-on
• The repo-owned skills live under openclaw_skills/ and are registered through skills.load.extraDirs in ~/.openclaw/openclaw.json.
• Terminal, launchd, and OpenClaw all share the same runtime path and config.

Metrics and History

Structured session events are written to:

~/.applypilot/metrics/session_events/YYYY-MM-DD.jsonl

Event types include:

• supervisor_started
• startup_delay
• wake_detected
• session_started
• session_finished
• session_failed
• control_action

Use applypilot report --section history or ./scripts/openclaw_activity.sh to inspect the recent ledger.

Local Command Backend

The fork includes scripts/local_apply_agent.py, a local Playwright-based apply agent that lets the command backend use either:

• a local Ollama server
• a remote Ollama server via OLLAMA_HOST or --base-url
• a remote OpenAI-compatible endpoint via --provider openai

The local agent now controls the browser for Workday and Greenhouse flows and supports:

• account sign-in
• account creation
• password-reset fallback
• optional IMAP polling for reset / verification emails
• Greenhouse verification-code retrieval over IMAP
• per-company ATS question memory and reuse
• command-backend silence timeouts so stuck agent runs fail fast instead of hanging forever
• ATS-aware prep/apply prioritization so easier supported targets can stay ahead of noisier tenants
• dry-run and live-submit result reporting back into ApplyPilot

Useful runtime tuning:

• APPLYPILOT_AGENT_TIMEOUT
• APPLYPILOT_AGENT_SILENCE_TIMEOUT
• APPLYPILOT_SUPPORTED_AUTOAPPLY_PATTERNS
• APPLYPILOT_PREP_PREFERRED_SITES
• APPLYPILOT_PREP_DEPRIORITIZE_SITES

Recommended small model for local development on low-memory machines:

• qwen3:4b for the best instruction-following / agent tradeoff
• llama3.2:latest as a zero-download fallback if you already have it

ollama pull qwen3:4b

export APPLYPILOT_APPLY_BACKEND=command
export APPLYPILOT_AGENT_COMMAND='python scripts/local_apply_agent.py --provider ollama --model qwen3:4b'

# Generate a prompt for a specific job and print the exact command to run
applypilot apply --gen --url URL --agent-backend command

# Or test the bridge directly with a saved prompt
python scripts/local_apply_agent.py --provider ollama --model qwen3:4b < /path/to/prompt.txt

Remote URL migration examples:

# Remote Ollama
python scripts/local_apply_agent.py --provider ollama --base-url http://REMOTE_HOST:11434 --model qwen3:4b

# Remote OpenAI-compatible server
python scripts/local_apply_agent.py --provider openai --base-url http://REMOTE_HOST:8000/v1 --model qwen3:4b

Mailbox automation example:

export APPLYPILOT_IMAP_HOST=imap.gmail.com
export APPLYPILOT_IMAP_PORT=993
export APPLYPILOT_IMAP_USER=you@gmail.com
export APPLYPILOT_IMAP_PASSWORD=your_app_password
export APPLYPILOT_IMAP_FOLDER=INBOX

If you keep a fixed password for one Workday tenant, scope it:

export APPLYPILOT_ACCOUNT_PASSWORD=your_existing_password
export APPLYPILOT_ACCOUNT_PASSWORD_HOSTS=thomsonreuters.wd5.myworkdayjobs.com

If different Workday tenants use different passwords, store them locally:

{
  "thomsonreuters.wd5.myworkdayjobs.com": "existing-tenant-password",
  "netflix.wd1.myworkdayjobs.com": "different-tenant-password"
}

Save that as ~/.applypilot/account_passwords.json, or point APPLYPILOT_ACCOUNT_PASSWORDS_FILE at another JSON file.

Question Memory

The local agent keeps ATS-specific company memory under:

~/.applypilot/question_memory/

Current structure:

• greenhouse/<Company>.json
• workday/<Company>.json

Each file stores:

• questions: reusable learned answers the agent can apply automatically later
• seen_questions: questions encountered on prior runs, with optional suggested answers for review

This memory is loaded after explicit host/company overrides and before global defaults, so curated overrides still win.

Quick inspection commands:

cd /Users/yashm/Documents/ai-job-application-agent
./scripts/show_question_memory.sh --list
./scripts/show_question_memory.sh --ats workday --list
./scripts/show_question_memory.sh "Ōura" --ats greenhouse

---

Contributing

See CONTRIBUTING.md for development setup, coding standards, and PR guidelines.

---

License

ApplyPilot is licensed under the GNU Affero General Public License v3.0.

You are free to use, modify, and distribute this software. If you deploy a modified version as a service, you must release your source code under the same license.