DigitalOcean is now offering a managed runtime for AI agents in public preview, allowing developers to deploy frameworks like Claude Code and LangGraph without configuring separate sandboxes. The service uses isolated Firecracker microVMs that preserve session state while pausing during idle periods to stop CPU and memory charges. These paused sessions resume in about 305 milliseconds, and a 2-vCPU agent averaging 25% utilization costs $0.06 per hour compared to $0.126 for full allocated capacity.

The platform integrates an Action Gateway with 16,000+ tools across 500+ providers, featuring a tool discovery system that matched intent with 99.3% accuracy in internal tests. Users can select from more than 75 open and proprietary models, while sensitive actions can be configured to require human approval. Early customer Qencode reported a reduction in support agent response times from hours to nearly instant.