open sourcelocal-firstApache-2.0free forever

The open source alternative to YC's Paxel.

Your AI coding sessions become a verifiable builder profile — open source and fully local. The same idea YC shipped as Paxel, built the way it should have been: your code, your prompts, your data, never uploaded. Yours forever.

Get your profile → Verify the privacy claim
$ git clone https://github.com/nextmillionai/nextmillionai && cd nextmillionai $ python3 -m nextmillionai start
exhibit 01 · the promisejun 6, 2026

"It runs locally inside Docker, and your code never leaves your machine."

@ycombinator, launching Paxel
exhibit 02 · the catchhours later

The community finds Paxel uploading prompt excerpts, file paths and git metadata to YC servers.

exhibit 03 · the cleanupsame day

The launch posts come down. The tool stays closed source, so you still can't check.

the community caught it — the full story →
92%of developers already build with AI toolsgithub developer survey
56%wage premium for proven AI skills, if you can prove thempwc ai jobs barometer, 2025
100%open source — every line that scores you is yours to read, fork, and runapache-2.0 · audit it yourself
the problem · everything still pretends it's 2015

Building changed forever. Proving it didn't.

AI rewired how software gets made. The way we show our work, get hired, and coordinate hasn't moved an inch.

your craft is invisible

The skill is the how. Nothing captures it.

How you prompt, steer, recover, and orchestrate agents, that's the craft. A CV flattens it to "2 years with LLMs." An interview tests what a JavaScript developer could show on pen and paper. Nothing, right?

what you actually are
Multi-Agent OrchestratorContext EngineerDebugging SurgeonRapid Prototyper
what the CV says→"AI engineer"
hiring reads facades

A green graph proves nothing now.

Pasted model output and real mastery produce the same commit. Keyword screens can't tell an agent orchestrator from someone who copies prompts, and the difference is everything.

two years of mastery
pasted model output
identical, to everything that exists today
coordination is ancient

Proof of work is finally effortless. We don't use it.

The eternal problem of human coordination was how hard it is to share proof of work. AI just made that free, and we still post jobs, cold-DM, and schedule interviews over six emails.

write a résumé nobody believes
post a JD, screen 200 keyword matches
six emails to schedule one call
your agent carries verifiable proof, agent-to-agent
why it's different · not another commit counter

Three reasons nothing else does this.

01

Built for a new craft.

Commits, résumés, and interviews were built for a world before agents. Building with AI is a different discipline, and this is the first instrument made to measure it, not a green graph with extra steps.

02

A map, not a ladder.

No percentiles, no leaderboards, no "top 10%." Every kind of builder is a craft, not a rung, scored for fit to how you actually work and never ranked against another person.

03

Open, not "trust me."

Paxel is closed — you couldn't see it collecting your data until someone reverse-engineered it. This one is open: read every line that scores you, run it offline, see exactly what leaves your machine.

see the proof ↓
run it once · five minutes, no account, your machine

The work GitHub never saw.

the how, measured

The commit was never the work.

GitHub saved the diff. Everything that produced it, how you prompt, steer, and recover, vanished. Run it once and the how becomes six measured dimensions, scored by plain arithmetic from your real sessions.

AI Craft Engineerdesigns the loopPrompt-Iterate mode
Signal Clarity
Build Stability
Decision Weight
Recovery Velocity
Context Command
Orchestration Range
the invisible 90%

The work that never showed up in a diff.

Most of your real work now lives in the conversation, not the commit. The tool counts it, starting with the one number no résumé and no green graph can show: the AI code that actually survived.

13,620
AI lines that survived in your codebase
github can't see a single one of them
4
parallel agents
72
subagents dispatched
252.7h
hands-on session time
11 days
ship streak
proof, not claims

Measured, cross-checked, hard to fake.

Every number is arithmetic from signals corroborated across your transcripts, your git history, and a local ledger, then stamped with its sources and confidence. Forging a profile means fabricating months of consistent logs across every tool.

✓verified via three independent sources
Claude CodeCursor IDEgit history
✓70 sessions · 2026-01-23 → 2026-06-11 · every claim traces to a measured pointer
assessment confidence83% · stated, never inflated
a new world order for work-proof

Yours to share. Anywhere, with anyone.

Sharing proof of work was always expensive, so we hire on proxies: résumés, interviews, take-homes. This is portable proof of work instead. Share a link, not your repo, and the code never leaves your machine.

nextmillionai.org/p/ada-democopy link
✓ derived signals only, never your code or prompts
✓ revocable any time, you own the switch
✗ nothing uploaded unless you publish, on purpose
a record that compounds

It grows as you do.

It keeps the history your tools forget and tracks how you're getting better over time. What's next: the same record reaching the people who should find you, and coordinating your fleet of agents.

↗ accelerating
soon

Reverse hiring & agent coordination — the right people reach your agent, and your fleet works from one shared record.

open vs. closed · the real difference

Open source. Read it yourself.

Paxel shipped closed — no one could see it uploading prompts and file paths until the community caught it. This is the opposite. The scoring, the methodology, and exactly what leaves your machine (nothing, until you choose to publish) are all there to read and run. You don't have to trust the pitch — check it.

# everything here is open to read and run yourself:
✓ the scoring — plain arithmetic you can audit, no model writes a number
✓ the methodology — versioned and public
✓ runs offline — the assessment never touches the network (and CI keeps it that way)
✓ publish — explicit, derived-only, revocable. never silent.
read the source → read the methodology → see the network rule →
open & living

Measuring how people build with AI is a new field that changes weekly — so the scoring isn't a fixed black box. It's a living methodology: open, versioned, and shaped by the community as the craft evolves. read it, challenge it →

where it's headed · the vision

The profile is the beginning.

Once how you build is legible and yours, everything broken about hiring and coordination becomes fixable, inside the agent you already use.

reverse hiring

Don't apply. Get approached.

The right people reach your agent, not your inbox. Anonymous until you decide otherwise.

agents

Agents that upgrade your AI game.

Tools that read your patterns and help you build sharper: tighter loops, fewer wasted sessions.

community

A million builders, learning openly.

Someone ships an app in five prompts; someone runs seven agents at once. New ways of building, shared.

your channel

Your agent is the channel.

From first hello to onboarding, the whole conversation runs through your Claude or Cursor.

the story, since you scrolled this far · receipts included

Why this exists.

I built this because no one should be able to take how you code, or your personal data. Then the idea got validated faster than I expected.

A
anshuli gupta ✓
@anshulix · 9:08 AM · Jun 6, 2026

you must be doing something right if Y Combinator had to copy your product and publish it as their own.

old video submitted to YC.

▶ intent AI — YCS26 application video, May 2026

Verified AI-skills profile · archetypes · composite score · agent-to-agent hiring, the product as pitched, one month before.

the application, uncroppedview on X →
Y
Y Combinator ✓
@ycombinator · Jun 6, 2026

"Today we're launching Paxel: a free tool that analyzes your Claude, Codex, and Cursor coding sessions and gives you a profile of how you build with AI.

It runs locally inside Docker, and your code never leaves your machine."

this post was later deleted
hours later: caught uploading prompt excerpts, file paths & git metadata
May 2026
Applied to YC (YCS26) with this exact product, profile, archetypes, score, agent-to-agent hiring, video included.
June 6, 2026
YC launches Paxel. The rejection email arrives the same day.
June 6, 2026 · hours later
The community finds Paxel uploading data it said stayed local. The launch posts come down.
today
You're looking at the version where that betrayal is structurally impossible.

The biggest accelerator on earth proved the market. We built the one that keeps the promise.

the full record — what the community said about Paxel →

Your work is already the proof. Make it legible.

$ git clone https://github.com/nextmillionai/nextmillionai && cd nextmillionai $ python3 -m nextmillionai start
★ Star on GitHub
made by · so you know who we are

Be known for how you build with AI.

This tool is the first piece of nextmillionai, a home for AI builders to own their craft: keep it, measure it, grow it, and one day be found for it, inside the agent you already use. Open source, local-first, and always free for developers.

★ Star on GitHub
The platform — reverse hiring, agents, a community — opens with sign-ups soon. Join the waitlist to be first.