SCROLL






2026

HuntOS

huntos

Local-first, agent-driven job application platform — SvelteKit, Mastra agents, CDP browser automation, SQLite, and self-hosted models (Ollama / LM Studio).

Back
Mini Map:
bloopchemical-pfdelderabusefuse-rusthuntosnew-portfolioold-portfolioremit-tracker

What it is

I built HuntOS as a local-first “career command center”: it automates the tedious parts of a job search so you can focus on interviews. I wrote a longer story on DEV about the vibe-coding / agent-driven workflow I used (design-first specs, Zed agents, Claude 3.5/Opus) and roughly 4 days of focused work in that narrative.

Capabilities

  • Job discovery — Scraping from boards such as LinkedIn, Greenhouse, and others, with filters for preferences (location, comp, role type, etc.).
  • Tailored resumes — Agent-generated Markdown and PDF resumes per job and template.
  • Browser automationChrome DevTools Protocol (CDP) drives real Chrome: navigates to forms, fills, and submits (with a dedicated Chrome instance you log into for sites like LinkedIn).
  • Pipeline — A Kanban-style view to track applications end to end.
  • Audit trailScreenshots and logs of agent steps for transparency and debugging.
  • Model routing — Cheaper models for easy tasks, heavier models for hard reasoning; support for Ollama and LM Studio to keep the stack self-hostable and cost-aware.
  • StorageSQLite for a fully local data layer (no required cloud DB).

Stack & license

  • Bun runtime, SvelteKit UI, Mastra for AI agents, CDP for automation, SQLite for persistence.
  • I license HuntOS under the GNU General Public License v3.0 (GPL-3.0). I chose it on purpose: I want the project to stay by individuals, for individuals—I don’t want someone to take my work, wrap it, and spin up a business selling it or charging for what should stay a community tool. GPLv3 is strong copyleft: if you distribute a modified version, the same freedoms (and obligations) follow. If you need something different for a serious commercial use, talk to me or use something else; don’t treat the repo as a product starter kit.

Notes from shipping

  • In testing I saw on the order of a ~65% successful submission rate; I still hit rough edges on login-only flows, strict validation, and the occasional model quirk (e.g. typing the wrong value shape into a field). The DEV post goes into that honestly.
  • Source: Blakeinstein / HuntOS on GitHub.