Why Katanemo is joining DigitalOcean
When we started Plano, our open source project for agentic systems, we believed the hardest part of building agents was not calling an LLM. It was getting agents into production quickly and reliably, then improving them over time without constantly wiring together glue code. Serving thousands of developers only strengthened that view. Again and again, we saw teams lose time to the same delivery problems around orchestration, observability, policy enforcement, and control, even though those concerns sit outside the core product logic developers actually want to build.
That conviction is also what makes DigitalOcean the right home for our next chapter. DigitalOcean has always excelled at turning complexity into simple, purpose-built primitives developers could trust: first with Droplets, then with Spaces and App Platform. The agentic era now needs that same instinct applied to AI infrastructure. Developers need more than raw GPU access. They need infrastructure primitives that make agentic systems reliable, observable, and easy to deploy in production. That is the essence of what we have been building here, and why joining DigitalOcean makes sense: together, we can help developers move from prototype to production with far greater simplicity, predictability, and confidence.
Where Plano and Our Model Research Fit
Plano is our framework-agnostic data plane for agentic systems, built to handle the traffic flowing to and from agents in production. It was designed as a purpose-built primitive with orchestration, safety, observability, and broad model access built in, so developers can stay focused on their agent’s core logic instead of the surrounding infrastructure. At DigitalOcean, that technology will evolve from a standalone open source project into a more native part of the platform, creating a much cleaner path from prototype to production.
Our model research sits alongside that same vision. We focused on how to deliver agents safely, efficiently, and with stronger real-world performance through work in model routing, governance and learning, and agentic performance. That includes compact orchestration models built to direct traffic between agents, preserve context across handoffs, and make multi-agent systems more reliable in practice. Together, Plano and our research are aimed at the same outcome: giving developers infrastructure and intelligence they can actually depend on in production.
And if you want to see where it’s all headed, come find us in person at Deploy: The Conference for the Inference Era.
We’re glad you’re along for it.
— The Katanemo Team