Maritime launches a managed cloud for production AI agents: each agent gets its own persistent microVM for $1 a month at 2GB RAM
Synopsis
Maritime released a managed cloud for production AI agents that places each agent in its own isolated, persistent microVM and layers on composable services for sleep/wake, persistent storage, triggers, scheduling, per-user routing and spend caps, health monitoring, failure recovery and zero-downtime updates. The team says companies still had to build this infrastructure themselves even after choosing a sandbox provider, and its cost work brings a 2GB-RAM agent to $1 per month with unlimited usage at that price.
Interpretation
Maritime turns the runtime for production AI agents into a managed cloud service, giving each agent an isolated, persistent microVM plus a configurable environment. After talking with a couple hundred companies building AI agents, the team says those companies still had to build additional infrastructure to make agents work even after picking a sandbox provider. The team groups the friction into agent lifecycle management, custom environments and cost optimization, and says sandboxes address roughly 10% of the infrastructure problem.
On top of compute, Maritime offers composable services for provisioning, persistent state, sleep/wake orchestration, triggers, schedule creation, monitoring, network configuration, updates, rollbacks, recovery, per-agent spend tracking and login handoffs. These are the sandbox management tasks the team says companies otherwise build and maintain themselves. The team says it spent substantial effort on cost, including scale-to-zero, copy-on-write over shared base images, compressed memory snapshots, hot/warm/cold snapshot and disk tiering, balanced against performance metrics such as latency.
Perspective
The service targets teams building production AI agents and covers managed capabilities from custom environments to lifecycle management; the launch asks users to report what they built on top of sandboxes and what they could not do with VMs.
The cost and friction judgments come from the team's synthesis of conversations with a couple hundred companies, and the pricing is the team's own; the launch gives no independent benchmarks or comparison against existing sandbox providers.
