One agent practice. Every customer stack.
Your customers are on Foundry, on Bedrock, and on their own clusters. Build the agent once, deploy it onto whichever they already bought, and put your name on the front.
promptarena deploy - Your own Kubernetes
Translated into resources your cluster already understands.
PromptArena-deploy-omnia - AWS Bedrock AgentCore
Ships a runtime that serves the agent over HTTP and A2A from inside AgentCore.
promptarena-deploy-agentcore - Vertex Agents
Deployed and managed as a reasoning engine on Google Cloud.
promptarena-deploy-vertex - Azure AI Foundry
A hosted agent in your own container, on per-session-isolated infrastructure.
promptarena-deploy-foundry
Four adapters, all public repos. Microsoft already gives us Foundry stops being an objection: they can have Foundry, and you still run one practice.
A reseller programmeprogram makes you a channel. This makes the practice yours.
Your customer, your brand, your engagement, your rate. We’re the platform underneath and we stay there.
Three of the four layers below belong to the customer. The one that’s ours is the one nobody pays a consultancy to build twice.
A four-layer stack. The top layer, the engineering lifecycle, is Omnia. The three below it are the customer’s: their PromptPack, their runtime, their Kubernetes.
Free until a customer is in production.
Free. Every feature, no clock.
Brand it, learn it, demo it. Self-hosted, so your only cost is the cluster it runs on.
One production-shaped environment. Free.
Built the way you’d build a customer’s. For demos and the reference implementation you point at.
One licencelicense per customer environment.
Standard published tiers. Whoever holds the paper, you or them.
Published prices, no partner discount. Free to build and sell is the concession. The tiers →
Your customer never pays per run.
Nothing is metered inside a licensed environment: agents, executions and tokens are all inside the number. When the thing you built for them takes off, it’s your engagement going well, not a new line on their bill.
Your customer buys the hardware. Buys the tokens. Then pays three cents every time an agent runs. Forever. On their own infrastructure.
Ten million runs a year is £230,000 in fees. Here it’s inside the £90,000.
Two things to know before you build on it.
Your brand goes on the console. Not on the cluster.
Rebranding the dashboard is shipped. It stops there, and their platform team will open a values file, so:
On their own cluster you get the lot. On the other three, most of it.
An agent on one of the other three is managed from the console the same way as one on their cluster. What stays behind is what that platform doesn’t offer, cell by cell, on Run anywhere.
Testing runs inside your customer’s boundary, and so does the simulation it runs against.
No programmeprogram, no directory, no certification. A handful of firms at a time, terms that hold, and the people who build it on the call.
Which three customers could you show this to next month?
If you have three names, there’s a conversation.