Daytona is building agent-native infrastructure to support large-scale, stateful agent workloads through fast, flexible computing environments designed for long-running execution.
Croatian-founded Daytona has raised a $24M Series A to build infrastructure designed for large-scale agent workloads. The round was led by FirstMark Capital, with Matt Turck joining the board. Additional participation came from Pace Capital, Upfront Ventures, Darkmode, and E2VC, along with strategic investments from Datadog and Figma Ventures. The round also includes a group of angel investors, including Gorkem Yurtseven (Co-founder of Fal), Theo Browne (Founder of T3 Chat), Eno Reyes (Co-founder of Factory.ai), Nikita Shamgunov (Founder of Neon), and others.
Every knowledge worker relies on a computer. As software agents take on more work, they will require computing resources at a much greater scale, potentially spanning millions of concurrent environments.
Most cloud infrastructure today is optimised for production workloads – stateless, immutable, and designed for consistent execution. While effective for serving software, this model is less suited to development and experimentation, which depend on flexible, stateful environments. Agents have similar needs but operate at far higher speed and scale, requiring environments that can launch in milliseconds, branch into parallel executions, support snapshots, and scale across large numbers of concurrent instances.
Daytona addresses these needs by introducing sandboxes as a core infrastructure primitive. A sandbox is a programmatic, composable computing environment in which CPU, memory, storage, GPU, networking, and the operating system can be provisioned on demand. These environments can be started, paused, forked, snapshotted, or terminated at any point during execution.
Founded in 2023 by Ivan Burazin (CEO), Vedran Jukić (CTO), and Goran Draganić (Chief Architect), Daytona focuses on providing programmable, sandboxed compute that allows agents to run code, explore alternative execution paths, and persist state at scale.
Using Daytona, an agent can launch a sandbox, run for extended periods, reach decision points, and fork into parallel branches to evaluate alternative approaches. Promising branches can be snapshotted, while others are discarded. State persists across failures, and execution paths can be cloned, resumed, or merged. Workloads may run for minutes or for days.
This model reflects a broader shift from cloud primitives designed around human workflows to infrastructure optimised for agents. Daytona focuses on making dedicated computing environments for agents practical through rapid startup, persistent state, and integrated tooling for activities such as writing code, using version control systems, and executing workloads securely at scale.
Following the Series A, Daytona plans to expand beyond sandboxes to support a broader set of agent-native infrastructure. The company will scale its systems for higher volumes of concurrent agent workloads, deepen integrations with developer and agent tooling, and continue improving reliability, security, and performance.
Daytona also plans to grow its team to support product development and customer adoption.