Large language model 2.1
Zurla Swarm 2.1
Reads advertising like language, answers in plain words and runs the AI agents in every auction.
Advertising has a language.
speaks it.
The first LLM for advertising.
Frontier superintelligence, built into the programmatic stack. Zurla Swarm 2.1 thinks: it plans your entire ad operation, from media planning to analysis. Its infrastructure acts: every integration, every media buy, every fraud check.
SSP, DSP, website owner, ad agency or app developer: if ads are your business, Zurla runs them.
Models
Three models.
One Zurla.
Each one does one job, and they work as one. You ask Zurla in plain words, and it runs on all three.
-
-
Vision model 2.4
Zurla Vision 2.4
Looks at pages, images, video, app screens and ads, classifies them and reads the intent in each image, so every ad lands in the right place.
-
Goal optimization model 3.5
Zurla Northstar 3.5
Turns a brand or performance goal into a route, decides on every impression in under 1 ms and keeps you on course.
Research
Teaching a model
the language of advertising.
Our research asks what it takes for a model to understand advertising, plan toward a goal and hand that plan to agents that act on it, safely, in auctions that finish in under 100 ms. These are the questions we work on.
Fig. 01
Reasoning about auctions
How the model weighs bids, floors, consent and brand rules, picks the outcome that serves your goal, and explains it in one plain sentence.
Open question
What makes an explanation of an auction decision clear enough to act on?
Fig. 02
Plain-language goals
Turning a sentence like “Grow revenue 15% by December” into a goal the model can plan toward, check every hour and report on.
Open question
When a goal leaves room for choices, what should the model ask before it acts?
Fig. 03
Forecasting
Traffic, demand and prices, months ahead, so publishers can sell the future and agencies can plan it.
Open question
How should a model say how sure it is?
Fig. 04
Agents in every auction
How thousands of agents coordinate inside every auction, all aimed at the same goals.
Open question
How do thousands of agents stay aimed at one set of goals?
Fig. 05
Judgment at auction speed
Turning a language model’s plan into decisions its agents make inside an auction that finishes in under 100 ms.
Open question
How much reasoning fits inside one auction?
Fig. 06
Measuring what matters
Judging the model on real results: every change tested against a control group, a share of traffic left as it was.
Open question
What should “better” mean for the advertiser, the publisher and the reader at once?
Zurla at work
AI agents
in every auction.
Thousands of agents, every millisecond, all working toward your goals. Every auction finishes in under 100 ms.
The thinking happens before the auction: Zurla Swarm 2.1 turns your goals into a plan, and Zurla Vision 2.4 classifies every page and ad ahead of time. In the auction itself, fast agents follow that plan. Zurla Northstar 3.5 prices each impression against your goals in under 1 ms, and the agents weigh every bid and pick the winner.
64ms
Auction runningAuction closed
1,284 agents on this auction
You: Why did that ad win?
Zurla: It bid the most and met every rule: consent given, a brand-safe page, known bots kept out. I closed the auction in 64 ms.
- 0 ms Request in
- 8 ms Request read
- 21 ms Consent and brand rules checked
- 39 ms Every bid weighed
- 64 ms Winner chosen · ad on the page
Illustrative data · one auction
Safety
Safe by design.
Tested first. Explained always. Undone in one click.
Zurla handles real money in real time. So every action is tied to your goals and kept inside your rules, and you decide what it may change on its own.
Zurla · Pellbrook Media
Why did sports revenue go up on Saturday?
I raised the floor from $1.80 to $2.20 before kickoff and tested it on 20% of sports pages first. It earned 9% more. Shall I roll it out to every sports page?
Yes, roll it out.
Done. It’s live on every sports page, with a note on what changed. Undo in one click.
Rolled out after your OK+9% on sports pages
Ask Zurla about any change
Illustrative data
Always on
- Every auction under 100 ms
- Native ads labeled
- Rewarded is opt-in
- Budget caps
- Brand safetyEvery page, app and ad is checked against your brand rules before a bid goes out.
- Fraud kept outKnown bots and invalid traffic are kept out before any buyer sees them.
- Privacy and consentEach person’s choices are read and respected: IAB TCF v2.3 in Europe and the UK, GPP for US state privacy choices.
- Tested firstEvery change runs on a share of traffic, against a control group, before it rolls out.
- Your approvalYou choose what Zurla may change on its own and what waits for your OK.
Careers
Join us.
Help build our LLM for advertising. Hard problems in real time: plans a model makes in seconds, carried out in auctions that finish in under 100 ms, and revenue your work helps grow.
See open roles