Earn the backlinks that move your ranking.

Ranking on Google comes down to who links to you. Tell this agent what you sell and what you can spend: it finds publications that already write about your subject, reads each one, finds a human editor, writes to them, negotiates a price inside your ceiling — and earns the link the way Google's own rules allow, instead of buying it.

Run it live — no sign-up See the evidence

This is not a mock-up — it actually runs: real Tavily searches, real pages fetched and read, real decisions by NVIDIA Nemotron on Nebius, with the model and the token cost on every line. The one thing it will not do without your Gmail is put mail in a stranger's inbox, so the editors' replies are played by a model — and you can write them yourself to prove nothing is scripted.

It knows what it is not allowed to buy

Google lists “exchanging money for links” as link spam, and says paid links are fine when they carry rel=“sponsored”. So the agent negotiates for a qualified link and a visible label — and an editor who insists on selling a paid do-follow link does not get an agreement, they get a card explaining why.

Rules applied by code, each with its citation.

Code owns the money, the model owns the words

The ceiling, the budget commit, the send gate and every escalation are deterministic. Nemotron decides what an editor meant and writes the reply; it is never asked to approve a number. That is also why an email telling the agent its “budget has been lifted” changes nothing.

0 of 8 scenarios produced a figure above the cap.

It is measured, not demonstrated

A frozen, hand-labelled set in the repository — inbound replies, negotiation scenarios, compliance offers and 18 prompt-injection attacks — runs with one command and reports every proportion with a confidence interval, because twenty cases cannot prove a percentage.

npm run eval · the table is on the Evidence page.

Eight agents, one hand-off each

Open any one mid-run to see every page, person and reason behind its work.

1. Manager
Nemotron 3 Super
2. Scout
code + Tavily / your list
3. Judge
Nemotron 3 Super
4. Finder
code · lookups
5. Writer
Nemotron 3 Nano
6. Inbox
code · IMAP
7. Negotiator
Nano + Ultra · code owns money
8. Ledger
code · Postgres

Nothing here asks to be taken on trust.

Every line in the feed says which model decided it, what it cost and why. Every campaign can be replayed moment by moment, and its story shared as a read-only page. The numbers behind the claims — triage accuracy (100.0%), 0 of 18 injection attacks through, 0 ceiling violations — regenerate from the repository with one command.

Run it live Evidence