LinkedIn Easy Apply, automated —
with your full review.

The algorithm, step by step

Every step lives in core/easy_apply.py. Open it in Notepad to verify — it's ~450 lines of commented Python.

1. 1

   Navigate to a filtered LinkedIn search

   We build a URL with your target titles, your primary market's cities, and four LinkedIn filters: f_AL=true (Easy Apply only), f_TPR=… (date posted: 24h / 7d / 30d), f_E=… (experience levels you target), sortBy=DD (newest first). Same query a human would type — just typed for you.

2. 2

   Lazy-scroll until ~25 cards are loaded

   LinkedIn lazy-loads results on scroll. We nudge the left rail 0.9 viewport-heights at a time, 4 times max, with 800ms between scrolls to mimic a human browsing. Never more aggressive than that.

3. 3

   Deduplicate against your local history

   Every job you've already applied to (or already attempted) is stored in your local easy_apply_log. Cards you've seen before are skipped instantly — no double-applications, ever.

4. 4

   Apply the description filters

   Open the right-rail. Drop the job if its description contains any EASY_APPLY_SKIP_KEYWORDS (default: us-citizen, security clearance, senior director), if EASY_APPLY_REQUIRED_KEYWORDS are missing, or if a 'X+ years required' line exceeds your EASY_APPLY_MAX_YEARS.

5. 5

   Click 'Easy Apply'

   Only if a button labeled exactly 'Easy Apply' is visible. We never click any other apply button (we don't auto-fill third-party career sites — too risky).

6. 6

   Walk the multi-step form

   For each visible input on the modal: read its label, canonicalise it (lowercase + strip punctuation), look up the answer in PATTERNS → cache → AI fallback. Type the answer into the right control (text / select / radio / checkbox / textarea).

7. 7

   Click 'Next' or 'Review' and repeat

   LinkedIn typically has 2–5 steps. We keep walking until we either see the 'Submit application' button or hit step 12 (hard ceiling — anything beyond that is probably broken).

8. 8

   (Dry-run by default) Stop at 'Submit'

   On the very first run, EASY_APPLY_DRY_RUN=true. We screenshot the form ready to submit and stop. You audit, then flip the flag once you trust the bot. With dry-run off, we untick 'Follow company' (avoids spam) and click Submit.

9. 9

   Wait 20–60 seconds. Repeat.

   Random jittered delay between every application. The hard cap (EASY_APPLY_DAILY_CAP, default 10) is checked at the top of each iteration — when it hits, the whole run stops and you get a summary in the dashboard.

How question answering works

Four layers — fastest and most deterministic first, AI last. 99% of questions never reach the AI.

1. Layer 1

   Pattern map (deterministic)

   30+ regex rules covering 99% of real Easy Apply questions: first/last name, email, phone, years of experience, work authorization, sponsorship, notice period, salary, cover letter, LinkedIn URL, city, country. Lives in core/easy_apply_questions.py — open it in Notepad and you can edit it.

2. Layer 2

   Q&A cache (per-user, per-question)

   First time we answer 'Do you have a security clearance?' we store the answer locally in easy_apply_answers. Every subsequent form uses the same answer — consistency, no drift, no AI re-rolls.

3. Layer 3

   Gemini Flash fallback (free tier)

   If we hit a question that doesn't match any pattern and your GEMINI_API_KEY is set, we send only the question text + a short profile blurb to Gemini. System prompt forces a short, plain-text answer. No résumé, no .env, no PII beyond what's already on your LinkedIn.

4. Layer 4

   'Needs review' fallback

   If both pattern AND AI fail (e.g. AI key missing, or the question is genuinely weird), we screenshot the modal and log the application as needs_review. Your dashboard surfaces it; you finish the application manually in 30 seconds.

How to enable it

1. 1. Set the flag in .env

   ENABLE_EASY_APPLY=true
   EASY_APPLY_DRY_RUN=true     # keep this true for the first week
   EASY_APPLY_DAILY_CAP=10
   LINKEDIN_COOKIE=AQEDAS...   # your li_at cookie

2. 2. One-time install (~150MB Chromium)

   .venv\Scripts\python.exe -m pip install playwright google-generativeai
   .venv\Scripts\python.exe -m playwright install chromium

3. 3. Run it (dry-run)

   EASY_APPLY.bat                 # Windows
   mac/EasyApply.command          # Mac

   Chromium opens. Watch the bot fill each form. Screenshots land in data/easyapply_screenshots/.

4. 4. When you trust it, flip the switch

   # Either: set EASY_APPLY_DRY_RUN=false in .env
   # Or:     run with the --no-dry-run flag
   .venv\Scripts\python.exe jobybot.py easy-apply --no-dry-run

FAQ