Skip to content

Bot Operations

Jump to bottom
AbhishekMandapmalvi edited this page Mar 11, 2026 · 1 revision

Bot Operations

Bot Lifecycle

The bot transitions through the following states:

         ┌──────────┐
         │          │
    ┌───▶│   IDLE   │◀──── stop / schedule end
    │    │          │
    │    └────┬─────┘
    │         │ start / schedule begin
    │         ▼
    │    ┌──────────┐
    │    │SEARCHING │───── querying LinkedIn / Indeed
    │    └────┬─────┘
    │         │ jobs found
    │         ▼
    │    ┌──────────┐
    │    │FILTERING │───── scoring + ATS detection
    │    └────┬─────┘
    │         │ qualified jobs
    │         ▼
    │    ┌──────────┐
    │    │APPLYING  │───── form filling + submission
    │    └────┬─────┘
    │         │ cycle complete
    │         ▼
    │    ┌──────────┐
    │    │ PAUSED   │───── waiting for next cycle / review
    │    └────┬─────┘
    │         │ next cycle / all reviewed
    └─────────┘
State Description
Idle Bot is stopped. No background activity.
Searching Querying job boards for listings matching search criteria.
Filtering Scoring found jobs and detecting ATS platforms.
Applying Generating tailored documents and submitting applications.
Paused Waiting between cycles, or waiting for user review in review mode.

Main Loop Flow

The bot's main loop (bot/bot.py :: run_bot()) executes this pipeline repeatedly:

┌─────────┐    ┌─────────┐    ┌─────────┐    ┌─────────┐    ┌─────────┐    ┌─────────┐
│ Search  │───▶│ Filter  │───▶│Generate │───▶│  Apply  │───▶│  Save   │───▶│  Emit   │
│         │    │ & Score │    │  Docs   │    │         │    │  to DB  │    │ Events  │
└─────────┘    └─────────┘    └─────────┘    └─────────┘    └─────────┘    └─────────┘

1. Search

  • Iterates over configured search_engines (LinkedIn, Indeed).
  • Each searcher queries with job_titles and locations from search criteria.
  • Returns RawJob objects (title, company, URL, location, description snippet).
  • Deduplicates against previously seen jobs (by URL).

2. Filter and Score

  • score_job() evaluates each RawJob against user criteria:
    • Keyword matching -- title and description vs. job titles and skills.
    • Location matching -- job location vs. preferred locations, remote preference.
    • Experience level -- entry/mid/senior alignment.
    • Excluded companies -- hard filter, score = 0 if matched.
  • detect_ats() identifies the application platform from the job URL.
  • Returns ScoredJob objects with a numeric score (0-100).
  • Jobs below the score threshold are discarded.

3. Generate Documents

  • If an AI provider is configured, calls generate_documents() to produce:
    • A tailored resume (rendered to PDF via ReportLab).
    • A tailored cover letter.
  • If no AI provider, falls back to template-based generation using profile data.
  • Reads all experience files from ~/.autoapply/experience_files/ for context.

4. Apply

  • In full_auto mode: proceeds immediately.
  • In review mode: adds to review queue and waits for user action (approve/reject/skip/manual).
  • In watch mode: skips application entirely.
  • Selects the appropriate applier from the APPLIERS dict based on detected platform.
  • Applier fills the form and submits with human-like behavior.

5. Save to DB

  • Creates an application record in SQLite with status, metadata, and event history.
  • Stores job description text for reference.

6. Emit Events

  • Pushes real-time updates via SocketIO to the dashboard:
    • job_found -- new job discovered
    • job_applied -- application submitted
    • job_review -- job added to review queue
    • feed_event -- general progress updates

Apply Modes

Full Auto

Search → Filter → Generate → Apply → Save

The bot operates completely autonomously. Every job that meets the score threshold is applied to automatically, up to max_applications_per_day.

Best for: Broad job searches where you want maximum application volume.

Review

Search → Filter → Generate → [PAUSE: User reviews] → Apply (if approved) → Save

The bot finds and scores jobs, then presents them in the review queue on the dashboard. For each job, you can:

  • Approve -- Bot submits the application.
  • Reject -- Job is discarded.
  • Skip -- Job stays in queue for later review.
  • Apply Manually -- Opens the job URL in the platform login browser for you to apply yourself.

Best for: Targeted searches where you want control over which jobs to apply to.

Watch

Search → Filter → Save (as "watching")

The bot finds and scores jobs but never submits applications. Useful for monitoring the job market, testing search criteria, or building a list of interesting positions.

Best for: Market research, criteria tuning, or when you are not actively job searching.

Supported Platforms

AutoApply can search on and apply to the following platforms:

Platform Search Apply Applier Class Key Technique
LinkedIn Yes Yes (Easy Apply) LinkedInApplier Multi-step Easy Apply modal
Indeed Yes Yes (Quick Apply) IndeedApplier Quick Apply flow
Greenhouse No Yes GreenhouseApplier Standard form fill (boards.greenhouse.io)
Lever No Yes LeverApplier Standard form fill (jobs.lever.co)
Workday No Yes WorkdayApplier Multi-step wizard (data-automation-id selectors)
Ashby No Yes AshbyApplier Single-page form (jobs.ashbyhq.com)

Greenhouse, Lever, Workday, and Ashby are ATS (Applicant Tracking System) portals. The bot reaches them when LinkedIn or Indeed job listings redirect to an external application page on one of these platforms.

ATS Detection

ATS detection uses URL fingerprinting defined in ATS_FINGERPRINTS (in core/filter.py). The bot examines the job application URL and matches against 9 known patterns:

URL Pattern Detected Platform
linkedin.com linkedin
indeed.com indeed
boards.greenhouse.io greenhouse
greenhouse.io/embed greenhouse
jobs.lever.co lever
lever.co/apply lever
myworkdayjobs.com workday
wd5.myworkday.com workday
jobs.ashbyhq.com ashby

If no fingerprint matches, the job is skipped (no generic form-filling is attempted).

Job Scoring

The score_job() function in core/filter.py produces a score from 0 to 100:

Factor Weight Description
Title match 40 How well the job title matches configured job_titles
Skill match 25 Keywords from skills found in the job description
Location match 20 Geographic or remote alignment
Experience level 15 Entry/mid/senior alignment with experience_level
Excluded company -100 Hard disqualification if company is in excluded_companies

Jobs with a score below the threshold (default: 30) are discarded without further processing.

Human-Like Automation

All appliers inherit from BaseApplier (bot/apply/base.py) which provides human-like interaction patterns to reduce detection risk:

Behavior Range Implementation
Typing delay 30-80ms per character _human_type() with random jitter
Click pauses 0.5-2.0s between actions _random_pause() with uniform distribution
Scroll behavior Gradual scrolling Page scrolls in increments
Captcha detection Per-page check _detect_captcha() pauses and alerts user

If a CAPTCHA is detected, the bot pauses and emits an event to the UI, allowing the user to solve it manually in the login browser.

Browser Session

AutoApply uses a persistent Playwright browser context managed by BrowserManager (bot/browser.py):

  • Profile directory: ~/.autoapply/browser_profile/
  • Session persistence: Cookies, localStorage, and login sessions survive between bot runs.
  • Single instance: Only one browser context is active at a time.
  • Separate from Electron: Playwright uses its own Chromium binary (installed via playwright install chromium), not Electron's bundled Chromium.

Login Flow

Before the bot can search or apply on LinkedIn/Indeed, you must be logged in:

  1. Click the platform login button on the dashboard (or use POST /api/login/open).
  2. A Playwright browser window opens to the platform's login page.
  3. Log in with your credentials.
  4. Close the login browser (POST /api/login/close).
  5. Your session is saved in the persistent context and reused for all bot operations.

Screening Answers

For ATS platforms that present screening questions (especially Workday and Ashby), AutoApply auto-fills answers from profile.screening_answers in your configuration:

{
  "screening_answers": {
    "Are you authorized to work in the US?": "Yes",
    "Do you require visa sponsorship?": "No",
    "What is your expected salary?": "$150,000",
    "Are you willing to relocate?": "Yes"
  }
}

The applier compares each screening question against the keys in your answers dictionary using fuzzy matching. If a match is found, the answer is filled automatically. Unmatched questions are left blank (and reported in the application event log).

You can manage screening answers through:

  • Settings UI: Profile > Screening Answers section
  • Setup Wizard: Step 2 (Profile) includes a screening answers sub-section
  • Direct editing: Modify config.json under profile.screening_answers

AutoApply Wiki

User Guide

Technical

Build & Deploy

Project

Clone this wiki locally