REGEN
Be the first application they read, not the 400th.
An open-source job engine that finds roles at the source, writes a resume that only claims what you can prove,
applies, and keeps a receipt for every submit. Your AI drives it. Zero model spend.
Live demo · How it works · Quick start · Wiki · Technical reference
The problem
A role goes live on a company's careers page. It reaches LinkedIn a day later, behind hundreds of identical resumes.
Auto-apply tools make it worse: one resume for every posting, invented answers for every gap.
What REGEN does
- Gets there first. Polls ~5,900 employer boards on 9 ATSs and six big-tech careers portals directly, rescanning every few minutes, so roles are caught the hour they go live. Discovery →
- Never bluffs. Every resume line comes from your Fact Bank; a validator rejects anything else. Tailoring →
- Skips what you'd lose. A fit gate reads the whole job description first: sponsorship, years, clearance, stack. Fit gate →
- Learns from no. Every rejection is read back and compared with the job descriptions that weren't rejected. Feedback →
- Proves it. A submit counts only when the confirmation page is saved. Evidence →
- Costs nothing to think. No model keys: Claude, Cursor or any MCP client operates it. Zero cost →
By the numbers
| 282 | 49 | ~5,900 | 43 | 0 |
|---|
| applications submitted | best day | employer boards watched | MCP tools | model API keys |
From the engine's own ledger as of 2026-10-22 (latest status per posting; python -m engine report lists the ones whose confirmation screenshot is still on disk). 261 tests in CI, plus a privacy gate on every push. Metrics →
How it works
discover ──► fit gate ──► compose ──► apply ──► inbox
▲ ▲ ▲ │
│ └──── rejection feedback ◄────────┘
│ │
sources ◄──────── MCP: your AI client ─┴──► gearbox (speed)
| |
|---|
| |
| Gearbox. One shift sets appliers and scan rate; the clutch couples what rejections taught. | The cord. 43 MCP tools reach every level, from finding new job sources to the kill switch. |
Maintenance (the engine's "oil change"). Every application leaves
leftovers: tailored resumes, job-description copies, verification-code and job-alert mail, closed postings, Docker cache.
The supervisor measures them every 5 minutes as a health score and clears them once a day (or on a click at /oilchange),
playing each step live and logging what each one removed. Proof is shrunk, never deleted; your Fact Bank, answers, application log and starred mail are never touched.
Everything runs in Docker on your machine: a pipeline, N appliers (headed Chromium), a supervisor that scales them,
a feedback loop on your inbox, a Postgres ledger, the control room, and the MCP server.
Architecture →
Quick start
git clone https://github.com/sameernagar-hub/regen-engine && cd regen-engine
cp -r profile.example profile # your facts and answers, once (git-ignored)
docker compose -f deploy/bank.compose.yml run --rm seed
REGEN_MCP_WRITE=1 docker compose -f deploy/bank.compose.yml up -d
Open the control room at http://127.0.0.1:3000 and plug your AI client into http://127.0.0.1:8790/mcp.
Step by step: Running the engine · Configuration · Technical reference.
Guardrails
Legal, sponsorship and EEO answers come only from your presets. CAPTCHAs and human checks always go to you.
Personal data is only typed into employer-owned ATS forms that pass a scam check. Your data stays on your machine;
the only things published are an anonymized activity feed and the code, and a privacy gate blocks names and secrets from both. Truthfulness and safety →
Learn more
Wiki · How it works · Architecture · MCP tools ·
Roadmap · FAQ · Changelog · MIT licensed
