An open execution platform for programmatic media.
Models make decisions in the bid stream. Always-on agents manage everything around them.
Bring your own, use ours, or combine both.
Media buying needs two very different kinds of intelligence.
Every impression requires a decision in milliseconds.
System 1 models run directly in the buying path and answer questions like:
These models are small, fast and purpose-built.
They do not need to reason about an entire campaign. They need to make one decision extremely well.
Campaign management happens over minutes, hours, days and weeks.
System 2 is a persistent agent that stays with the campaign throughout its life.
It can:
Use the agent built into the platform, or bring your own.
The entire media platform is an API.
Agents are native operators.
You should not need to live inside another adtech dashboard just to manage a campaign.
Connect your own Claude, ChatGPT or agent environment and tell it what you want.
For example:
Check this campaign every weekday morning. If CPA moves more than 15%, investigate why. Make small budget adjustments automatically. Ask me before changing models or increasing total spend.
Or:
Test the new store-visit model against the current model. Keep 80% of spend on the existing model. Move more traffic only if the new model performs better.
Schedules, guardrails and approvals remain explicit.
The agent keeps running even when you are not talking to it.
We are building an open family of models specifically for media decisioning.
Think of it as the beginnings of a Hugging Face or PyTorch ecosystem for programmatic decisioning.
Initial model families include:
Every model comes with the code, training recipe, evaluation approach and deployment format needed to actually use it.
Open models are only the starting point.
Use the platform to train or fine-tune them against your own campaign and outcome data.
The workflow is designed to cover the full model lifecycle:
You can compare models against historical auctions before putting spend behind them.
Your data and trained models remain yours.
There should not be one universal media algorithm.
Different models can contribute different pieces of the decision.
For example:
Different participants can contribute different models.
A model might belong to:
The platform provides a common way to run them together.
Models are portable assets.
The platform handles how those models are discovered, combined, evaluated and executed.
The intelligence belongs to whoever created it.
The core auction server is open source.
Developers can inspect the code, run it themselves, contribute to it or build their own infrastructure around it.
Our managed platform runs the same open core.
Buying logic should be portable.
A model or bidder can run close to the auction rather than every opportunity travelling through layers of external infrastructure.
The same workload can move between supported execution environments without being rewritten.
Containers make that possible.
They also make ownership clear: your logic is an actual deployable piece of software, not a setting hidden inside someone else's platform.
Every decision should be inspectable.
You get access to:
The logs are yours.
There is no additional fee to access your own event history or export it elsewhere.
Programmatic infrastructure processes enormous volumes of data.
That makes efficiency matter.
We want the real-time path to do as little unnecessary work as possible.
Those goals reinforce each other.
The core execution layer is designed around that principle from the beginning.
We want a clear path between buyer and publisher.
That does not mean every publisher must integrate directly with us.
It means avoiding unnecessary hops, reselling and duplication wherever possible.
Every impression should have understandable provenance.
For developers, one way to think about the platform is:
A fast execution environment.
An open model ecosystem.
And an always-on system capable of operating campaigns on top.
The aim is not to build another closed media platform.
It is to create infrastructure that agencies, advertisers, publishers and developers can build their own media systems on top of.
Join the waitlist