OrbitApply deploys a coordinated fleet of 8 AI agents to find jobs, research companies, tailor your resume, write cover letters, and track every application — entirely on your own machine. No cloud. No subscriptions. No data leaves your device.
The numbers are stark. General-purpose AI doesn't solve the coordination problem — OrbitApply does.
ATS systems filter out three-quarters of applications automatically. A general-purpose LLM has no idea how a specific ATS actually parses your file — OrbitApply's TAILOR agent runs an internal ATS simulation targeting 70+/100 before a single application goes out.
100 applications at 45 minutes each. An LLM cuts time per application but doesn't solve the coordination problem — you still have to find jobs, manage the tracker, prep for each company, and remember who said what when.
ChatGPT and Claude default to recognizable house styles. Experienced recruiters now spot generic AI output immediately. OrbitApply tailors to your voice from your own resume — not a template.
Using an LLM means starting from scratch every session. OrbitApply maintains your profile, tracks every application, schedules follow-ups, and triggers interview prep automatically when a stage changes.
The moment you mark a role as "Phone Screen" or "Interview," COACH activates automatically — generating a full prep pack with STAR answers, salary negotiation scripts, and company-specific talking points.
ORBI orchestrates every step in sequence, streaming live progress to the UI. Each agent hands off structured data to the next — no manual steps in between.
Reads your profile to build 15 targeted search queries across Lever, Greenhouse, Ashby, Workday, and SmartRecruiters. Runs all queries in parallel. Scores every result locally with a pure JS engine — no AI call, no API cost. Jobs scoring below 40 are silently dropped.
For each qualified job, RECON builds a structured intel profile: culture, funding stage, tech stack, Glassdoor signals, salary benchmarks (low/mid/high), red flags, and opportunity score. Saved and passed to TAILOR for personalisation and COACH for interview prep.
Runs before every TAILOR and SUBMIT action. Enforces daily budget cap ($5), apply limit (15/day), blacklist, and routes sensitive fields (salary, diversity questions, security clearance) to a human review queue. Cannot be bypassed by any agent.
Two sequential Claude Sonnet calls per job: first tailors your resume to mirror JD language and inject ATS keywords, then writes a 250–350 word cover letter. Saves named documents to Apply/applications/{Company} — {Job Title}/.
Registers every tailored application with full metadata, sets a 7-day follow-up reminder, and maintains running stats: total applied, response rate, daily count, and budget spent. No database — flat JSON files, fully on your machine.
Fires the moment you mark an application as Phone Screen or Interview. Generates a full prep pack: 10 STAR behavioural questions, 5–10 technical questions, salary negotiation script anchored 15–20% above the RECON benchmark, and 3 recent news items to reference.
Each agent has a defined role, model, and contract — no general-purpose prompting.
You need Node.js v18+, pnpm, and two API keys. That's the entire setup.
git clone from GitHub and cd into the project directory.
Copy .env.example to .env. Add your Anthropic key (sk-ant-...) and Tavily key (tvly-...). Both are required.
Create memory/profile.json with your target roles, salary range, skills, and location. Add your resume to memory/resume.md.
pnpm install && pnpm dev. Open http://localhost:3000. Click Run. Watch 8 agents go to work.
API Keys needed:
• Anthropic — console.anthropic.com (for all AI writing)
• Tavily — tavily.com (for job search & company research)
20+ years running large-scale enterprise transformation programs — change management, multi-country system overhauls. Over the last two years, 2,500+ hours building AI applications, agentic workflows, and automation systems entirely through prompt engineering. OrbitApply started as a tool built for a friend going through a broken job search process. When it helped, I cleaned it up, made it free, and put it here.