Zurla Swarm 2.1

Large language model 2.1

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.

Book a demo See it at work
AI agents in every auction
Thousands of agents, every millisecond, all aimed at your goals.
1,284agents working right now

Use cases

One model.
The whole job.

From building workflows to planning media, running programmatic operations and writing reports. Ask in plain words, and Zurla does the work.

Zurla Swarm 2.1Calderwick · Velstead Motors
  1. A Calderwick planner drops four files into the chat: Velstead Motors’ launch brief, brand guidelines, 24 creatives and last year’s results, and asks for a programmatic plan.
  2. Zurla reads them: the brief’s goal (3 million US car shoppers, November 3 to 30, $250,000, 70% in view), the brand’s colors, tone and rules, every creative, and the programmatic traffic for car shoppers.
  3. Zurla writes the plan as an artifact: KPIs, audiences, budget by channel, six inventory sources, flighting, bidding, creative mapping and brand safety, forecasting 3.2 million car shoppers with 74% in view.
  4. The planner approves it. Zurla creates the campaign and five line items, six private deals, uploads 26 creatives, sets targeting, brand safety and pacing, serves test impressions and schedules the launch.
  5. A week later Zurla reports: reach 1.1 million, ahead of plan, 76% in view and 69% completed views, with the changes it made.
Zurla Swarm 2.1Calderwick · Creative ops
  1. Calderwick’s creative ops lead describes a process once: every time a client sends creatives, check them against the specs and brand rules, make the missing sizes, get the client’s OK, then traffic them to every live line item.
  2. Zurla builds it as a workflow in eight steps: creatives arrive; check specs and brand rules; a branch, where anything that doesn’t pass goes to Fix it (add the label, re-export, or ask the client) and comes back; make the missing sizes; client approval; traffic to every live line item, tested on 10% first; and a summary for the team.
  3. A test run with Tidebolt’s 14 creatives: 12 pass, 2 are fixed, 36 sizes are made, Tidebolt approves all 50 in one click, and they go live on 9 line items.
  4. The lead turns it on for six clients. In its first week it runs 23 times: 318 creatives, 96 sizes made, 11 fixed before review, every one approved before it went live.
  5. The lead adds a step in plain words, a preview PDF before the approval, and Zurla puts it in as step 6.
Zurla Swarm 2.1Orrinth Exchange · Ad ops
  1. Orrinth Exchange’s ad ops lead asks Zurla to add Vellacourt as a new SSP, for video, connected TV, display and native, from the US and Europe, and to set up its RTB integration.
  2. Zurla creates Vellacourt SSP (partner ID VLC-SSP-01, starting at 20,000 QPS, Orrinth’s brand safety rules, TCF v2.3 and GPP), then sets up the OpenRTB 2.6 integration with three endpoints, US East, US West and EU West, and tests it: 1,000 test requests, all valid, answered in 34 ms.
  3. Zurla offers to email the details from the Gmail integration. The lead says yes, and Zurla writes and sends the email to Vellacourt’s ad ops team with the seat ID, endpoints, specs, test results and an integration guide.
  4. The lead asks to raise Vellacourt’s QPS from 20,000 to 60,000. Zurla does it in three checked steps, 30,000, 45,000 and 60,000, answering in 31 ms with 99.8% valid requests.
  5. The lead asks for the fill rate of Corrandel Bidder and Ostlyn DSP on Vellacourt’s traffic over 7 days. Zurla compares 418 million bid requests: Corrandel fills 87.4%, Ostlyn 81.2%, with Ostlyn ahead on native.
  6. Zurla makes the report: head to head, fill rate by day, by format and by hour, share of wins (56% and 44%), response times of 36 and 42 ms, and three next steps.
Zurla Swarm 2.1Pellbrook Media · Revenue
  1. Pellbrook Media’s revenue lead asks for October’s report for the board: revenue, where it came from, and what’s next.
  2. Zurla first matches every number to its source: 812 million impressions from the Zurla SDK, 486 million from the programmatic integration, 2,316 signed orders and 196 million pageviews. October’s revenue, $2,841,320, matches to the cent across all four.
  3. The report: October 2026 was a record month. Revenue $2.84 million, up 18% from September; $14.50 per 1,000 pageviews; 96% fill rate; 79% in view. Revenue by day, by source (SDK native 46%, programmatic 32%, SDK banner 22%), by section and by advertiser, led by Velstead Motors; what it means in plain words; and November’s plan, on track for $3.1 million.
  4. The lead clicks Oct 14, the strongest day, and asks why it was so strong. Zurla answers on the chart: the City Marathon live page drew 3.1 million pageviews, four times a usual Wednesday; Zurla opened it to every buyer at 6 a.m. and added a native slot, and that one page earned $71,200. Zurla adds it to the report.
  5. Zurla schedules the report for the board, 5 people, at the start of every month (first business day, 8 a.m.), with every number matched again before it sends. October’s report goes out now; November’s on Dec 1.
Zurla Swarm 2.1Calderwick · Velstead Motors
  1. Calderwick’s account lead asks for Velstead Motors’ November results deck for Friday’s review, the spreadsheet behind it and a one-page summary for their CMO.
  2. Zurla reads 41 days of campaign data, Velstead’s brand book and October’s deck, and makes all three side by side: an 8-slide deck in Velstead’s colors, a 4-tab sheet with every number sourced, and a one-page summary.
  3. The deck: November results; 3.4 million car shoppers reached, 113% of the goal; 76% in view and 69% completed views, both above goal; results by channel; weekday mornings drove twice the completed views; three changes, each tested on 10% first.
  4. The lead asks to make slide 4 about connected TV, with a chart instead of the table. Zurla redraws it: connected TV led the launch, with 91% of its ads watched to the end.
  5. Zurla exports the deck, the sheet and the summary and shares them with 4 people at Velstead, and will update all three when December’s numbers land.
Zurla Swarm 2.1Orrinth Exchange · Operations
  1. Orrinth Exchange’s operations lead asks Zurla to set up alerts and actions for its SSPs, DSPs and campaigns through Slack, Twilio, Salesforce and Mailchimp. Zurla connects the four tools.
  2. Zurla writes 12 rules. SSPs: past 90% of the QPS cap, raise it in checked steps and post to Slack; a new SSP goes live, create the Salesforce account and start Mailchimp onboarding; a contract renews in 30 days, open a renewal task. DSPs: a new DSP connects, open the opportunity and send a welcome; $50,000 spent in a day, text the account lead; win rate past 40%, offer more supply. Campaigns: a goal reached, share it and update Salesforce; 20% of the budget left, text the account manager; a launch, confirm it in Slack. Across the exchange: a traffic peak forecast, scale QPS ahead and tell on-call; bot traffic kept out, log it; every Monday, email each partner their results.
  3. The lead turns the 12 rules on. On Monday they handle 214 events with 388 actions, every one delivered: partner emails, Slack posts, a Salesforce opportunity, texts to account leads and a scale-up ahead of Friday’s peak.
  4. The lead adds a rule in plain words: when a campaign beats its goal by 10%, text me and post it to #wins. Zurla creates rule 13 and test-runs it: two campaigns match, the text and the Slack post go out.

Illustrative data · fictional brands

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.

  1. 0 ms Request in
  2. 8 ms Request read
  3. 21 ms Consent and brand rules checked
  4. 39 ms Every bid weighed
  5. 64 ms Winner chosen · ad on the page

Illustrative data · one auction

Zurla Swarm 2.1

Specs

Type
Large language model, made for advertising
Reads
Your words, ad requests, pages and app screens, bids, consent signals, goals and brand rules
Gives
Answers and reports in plain words, plans, the rules its agents follow in every auction, and a note on every change
Speed
Answers in seconds; its agents carry out the plan in auctions that finish in under 100 ms
Where you use it
In chat with Zurla, and in every auction: Zurla’s SDK, Prebid and RTB infrastructure
Works with
Zurla Vision 2.4 and Zurla Northstar 3.5
In Products
Swarm, AI agents in every auction, is built on Zurla Swarm 2.1

The Zurla model family

Three models.
One Zurla.

Each model does one job, and they work as one. Zurla Vision 2.4 sees the content, Zurla Northstar 3.5 sets the course, and Zurla Swarm 2.1 talks with you and acts in every auction.

Put Zurla Swarm 2.1 to work.

Book a demo, and see it read your auctions and answer in plain words.

Book a demo Introducing Zurla Chat