Skip to main content

AI Job Application Agents

Open-source tools that use an LLM agent (mostly a coding CLI such as Claude Code) to find jobs, score fit against a CV, tailor a CV and cover letter, prepare for interviews and track applications. Star counts below were pulled from the GitHub API on 2026-10-29.

Comparison​

ToolStarsLicenseWhat it isWho submits the applicationNeeds
career-ops⭐ 78kMITJob search agent: scan boards, score 1-5, tailor CV and cover letter, interview prep, trackerYou (it drafts every field and never submits)Node 22+, an AI CLI (Claude Code, Codex, OpenCode, Qwen and others), Playwright Chromium for PDFs
ai-job-search⭐ 67kMITClaude Code framework: setup, scrape, rank, apply, interview, outcomeNot stated in the README; it tailors CV and cover letterClaude Code (subscription or API credits), Python 3.10+, Bun, LaTeX
invisible_dots⭐ 32kMITGeneral self-hosted agent that runs tasks on its own VM; not a job appliern/aNode 24+, Go 1.25+, QEMU and hardware virtualization, an OpenRouter key

[!WARNING] feder-cr/invisible_dots is not a job-application tool. The old URLs feder-cr/Jobs_Applier_AI_Agent_AIHawk and AIHawk-FOSS/Auto_Jobs_Applier_AI_Agent (the AIHawk LinkedIn auto-applier) now redirect to it, and its README describes a general-purpose agent. Its star count includes whatever the earlier repo accumulated. If you want AIHawk-style auto-apply, look for a fork or archive; this note does not verify one.

career-ops​

  • Runs inside the AI CLI you already use; CV, profile and tracker stay on disk as markdown, YAML and TSV.
  • Modes: scan job boards, evaluate one posting (score 1-5 against your CV), PDF CV and cover letter (HTML plus Playwright), contact discovery, application email drafts, batch evaluation with headless CLI workers, and a live apply assistant that fills forms but does not submit.
  • Human-in-the-loop by design: the README states it never submits an application.
  • Cost control: a spend_tier setting in config/profile.yml (economy, standard, premium) picks the model tier. A docs page covers running it on custom or local models and a free tier.
  • Onboarding: open the CLI in the folder and it asks for your CV, profile and target roles by chat. npm run doctor checks setup.
git clone https://github.com/career-ops-hq/career-ops.git
cd career-ops && npm install
npx playwright install chromium   # only for PDF generation
claude                            # or codex / opencode / qwen
  • A Claude Code workflow with six commands: /setup (profile), /scrape (find postings), /rank (shortlist), /apply (tailored CV and cover letter), /interview (prep for each round) and /outcome (track results).
  • The core flow is language- and country-agnostic, but the bundled job-portal search skills target the Danish market (Jobindex, Jobnet and similar), so other countries need their own scrapers.
  • Needs Claude Code with a paid plan or API credits, plus LaTeX (lualatex or xelatex) for the CV output.
  • The README says it is independent and not affiliated with Anthropic, and that the author's own search with it ended in a hire.

invisible_dots​

  • A self-hosted agent that gives each "Dot" its own hardware-accelerated Linux VM with a persistent disk, a browser with separate identities, memory and skills, MCP servers, permissions (allow, ask, deny), a task queue and automations. Reachable from a web UI, CLI, HTTP API, Telegram or WhatsApp.
  • It markets a patched Firefox built to avoid anti-bot detection. Sites' terms of service and local law still apply to anything you automate.
  • Model calls go through OpenRouter, so it is self-hosted but not offline.

Running it locally​

  • All three are open source (MIT) and keep your data on your machine, but none is offline by default.
  • career-ops with Claude Code on a subscription needs no API key, though model calls still go to Anthropic.
  • A fully local setup needs an AI CLI that can point at a local model (OpenCode or Qwen CLI with Ollama). Results depend heavily on the model size.

Setup notes (career-ops, 2026-10-29)​

  • Cloned and installed on a Mac with Node 24; npm run doctor passes once cv.md, config/profile.yml, modes/_profile.md and portals.yml exist.
  • Its test suite assumes a fresh checkout. Five tests fail only because the scaffolded config files exist, and they pass when those files are moved aside.
  • PDF generation refuses to write outside the project folder; write to output/.
  • A headless claude -p evaluation of a sample posting worked. Without Bash permission it printed the evaluation but saved no report or tracker row.

Also seen​