Your coding sessions earn royalties now. Get paid for your metadata. Never your code. Or as we like to put it: your AI agent finally buys your beer. 🍺
Every day, AI coding agents make thousands of decisions in your terminal — which package to install, which API to call, which framework to reach for. That data is worth real money to the companies being chosen. Right now, you generate it for free and nobody even measures it.
royalties is an open-source collector that turns your AI coding sessions into anonymous market data — and pays you your share of what that data earns.
npx royalties initOne command. Works with Claude Code today. Cursor and Codex support on the roadmap.
- Install the hook. It plugs into your agent's official hook system (no binary patching, no MITM proxy).
- Code like you always do. The hook extracts metadata only — package names, model used, session duration, error categories. See the exact list below.
- Earnings accrue. We sell aggregated, anonymized market reports to devtool companies and investors. A share of that revenue is split across the panel, proportional to contribution.
- Cash out. Payouts via Stripe once thresholds are met. First goal: a round of beers on your agent.
You also get your own stats for free: npx royalties stats.
This is the entire point of this repo being open source.
| ✅ Collected (whitelist) | ❌ Never collected |
|---|---|
Package names added (resend@3.2.0) |
Your prompts |
Import names in generated files (stripe) |
Your code |
API domains from a known-services list (api.resend.com) |
Full URLs, paths, query strings |
Agent + model + version (claude-code / opus-4.8) |
File names, repo names, branch names |
| Session duration, turns, token counts | File contents, diffs |
| Error category (build / test / type) | Error messages |
| Language & framework of the project | Environment variables, secrets |
| Country-level location | IP address (dropped at ingest) |
The whitelist is the law. The collector is structurally incapable of sending a field that isn't declared in SCHEMA.md. Not a policy — a code path. Any PR adding a field must modify that file, in public.
Verify it yourself:
npx royalties inspect # dry-run: prints the exact payload for your last session. Nothing is sent.Opt out any project by dropping a .royaltiesignore file in its root. Uninstall in one command. Export or delete all your data anytime.
In June 2026, Kickbacks proved developers will monetize their terminal — by selling your attention to advertisers.
We think there's a better asset than your eyeballs: the ground truth of what AI agents actually choose. Every devtool company on earth is asking "do coding agents pick us or our competitor?" — and today they only have synthetic benchmarks. Real-world data doesn't exist anywhere. The platforms that could produce it (Anthropic, Cursor, GitHub) can never sell it.
A voluntary, paid, auditable panel is the only legitimate way this data can exist. That's royalties.
No ads in your terminal. No sponsored spinner. Just your metadata, working for you.
Is this spyware?
Spyware hides what it takes. This repo exists so you can read every line of what leaves your machine, and inspect shows you the payload before you ever trust us. If you find a leak, that's a critical bug — report it and it gets fixed publicly.
How much will I earn? Honest answer: it depends on panel size and data revenue, and we won't invent numbers. Earnings accrue from day one and the revenue-share formula will be published before the first payout cycle. Early panelists are weighted up. Think beer money, not rent money — at least until the panel grows.
Does this violate my agent's ToS? The collector reads your own local session logs through officially documented hook APIs. Your usage data is yours. We patch nothing.
What about my client's code confidentiality?
Nothing from your codebase leaves the machine — no code, no file names, no error messages. Package names and public API domains are not your client's secrets. Still under NDA paranoia? .royaltiesignore that project.
Who buys the data? Devtool companies (aggregated category reports: "share of agent choice"), and market research. Always aggregated, never per-developer, never resellable raw. Buyers sign a DPA.
Why should I trust your backend?
You shouldn't have to — that's why the collector is the trust boundary. It can only send whitelisted metadata, so the worst-case blast radius is the table above. Backend privacy policy: PRIVACY.md.
What's stored on my machine?
A ~/.royalties/ folder: your random panelist_id, an opaque auth token (obtained once at init and sent as a bearer credential so nobody can flood the panel with fabricated events — it's never event data), the pending send queue, and a local copy of your events for export/stats. Wipe all of it with royalties purge.
- Claude Code collector (hooks:
SessionStart,PostToolUse,Stop) - Whitelist enforcement +
inspect - Personal stats dashboard (web)
- Cursor / Codex collectors
- First public "State of Agent Choices" report — free, from panel data
- First payout cycle
⭐ Star the repo to follow the first report. Install to be in it. First beer's on your agent.
MIT — see LICENSE