Job Search OS

Personal agentic job-search operating system for reviewing opportunities, maintaining search profiles, generating truthful application materials, and keeping the final application decision under human control.

What This Is

Job Search OS is a protected single-user production app. It combines job discovery, evidence-backed resume and cover-letter generation, application tracking, local browser assistance, email outcome monitoring, Slack operations, and agent quality loops into one local-first workflow.

The app is built around a conservative operating model:

  • LinkedIn is treated as a lead, analytics, and public-content channel, not a scrape or auto-apply source.
  • Application submission, employer contact, email sending, calendar writes, and unreviewed LinkedIn publishing stay manual or explicitly approval-gated.
  • Jolene and the skill system classify actions as read_only, proposal, safe_internal, guarded_mutation, or external_blocked.
  • Deployed cron and sync endpoints must be protected with bearer secrets.
  • /api/system/health reports database readiness, stale work, required secrets, provider configuration, and worker readiness.

The canonical product roadmap and audit package lives in plans/FIRST_CLASS_AGENTIC_JOB_SEARCH_OS_AUDIT_ROADMAP_PLAN.md.

Quick Start

Install dependencies:

npm install

Start the local database, run migrations and seed data, then launch the app:

npm run dev:local

Open:

http://localhost:3000

DATABASE_URL comes from .env. The checked-in example points at Docker Postgres on localhost:5433; keep .env pointed at any existing populated local database you already use.

For manual setup, Docker profiles, and environment variables, see wiki/Getting-Started.md.

Verification

Run the standard local checks:

npm test

npm run lint

npx tsc --noEmit --pretty false

npm run build

Smoke and browser acceptance checks require a running app:

npm run smoke:pages

npm run test:e2e

Core Capabilities

  • Job discovery and scoring - searches direct ATS/company sources, broad open-web sources, and review-only lead channels, then dedupes and scores matches against enabled profiles.
  • Apply Workspace - the canonical review surface for tracked applications, packet readiness, direct application links, blockers, materials, answers, fit, research, history, and job details.
  • Resume and cover-letter generation - creates ATS-readable, evidence-backed materials with material-quality checks and export endpoints for manual review.
  • Jolene Chief of Staff - provides in-app operational help, app-state answers, delegated internal proposals, career sprint briefs, and confirmation-gated repair actions.
  • Email Ops - classifies job-response mail, matches messages to applications, extracts outcomes or interview signals, and routes ambiguous work for approval.
  • LinkedIn content and analytics - drafts public-safe build-in-public posts, supports approved publishing when connected, and tracks aggregate post analytics.
  • Agent quality and observability - records agent runs, examples, evaluations, improvement proposals, learning impact, and review gates before expanding automation.
  • MCP and integrations - exposes local tools through an MCP server and supports Chrome capture, GitHub context, notifications, Slack operations, and provider-backed AI.

Documentation Map

Intent Start here
I want to run the app wiki/Getting-Started.md
I want to understand the product wiki/Home.md and docs/USER_GUIDE.md
I want to configure integrations wiki/MCP-and-Integrations.md and wiki/Operations-and-Configuration.md
I want to understand agents/workflows wiki/Agents-and-Workflows.md and wiki/Command-Center-and-Jolene.md
I want to operate or debug it wiki/Operations-and-Configuration.md
I want the full user guide docs/USER_GUIDE.md

Main Surfaces

  • /dashboard - Today cockpit for search, review, apply, follow-up, blockers, and operating brief.
  • /jobs - exception and review queue for discovered jobs.
  • /applications and /applications/[id] - application tracking and Apply Workspace.
  • /resumes and /resumes/profile - resume sources, profiles, variants, uploads, and generated materials.
  • /evidence - candidate evidence and LinkedIn recommendation import.
  • /sources - source management, company targets, broad query coverage, and supported boards.
  • /agents - agent quality gates and reliability controls.
  • /settings - providers, notifications, automation policy, profile links, learning, and system configuration.
  • /linkedin-content - prompt-first LinkedIn content studio.
  • /architecture - generated system architecture report.

Operating Boundaries

The system can prepare, explain, score, draft, repair, and route work. It does not silently submit applications, bypass CAPTCHA, scrape LinkedIn jobs, publish LinkedIn posts, send email, contact employers, or mutate external calendars.

Local browser automation is assistive only: it fills safe fields, uploads prepared materials when visible, learns from manual edits, and stops before final submit. For details, see wiki/Application-Automation.md.

Generated materials must stay grounded in approved evidence. Resume, cover-letter, packet, and material-claim behavior is documented in wiki/Evidence-RAG-and-Materials.md.

Stack

  • Next.js App Router and React
  • TypeScript
  • Prisma with PostgreSQL and pgvector
  • MUI and Emotion
  • Vitest, Playwright, and ESLint
  • OpenAI structured outputs with deterministic fallbacks
  • LangGraph, LangSmith-compatible observability, ADK opt-in control-plane support
  • Slack Bolt, MCP SDK, IMAP/OAuth email sync, Chrome extension support

License

This project is licensed under AGPL-3.0-only. See LICENSE.