Maritime

Maritime is managed cloud infrastructure for AI agents. Its core model gives each agent a persistent isolated runtime with durable storage, sleep/wake lifecycle management and multiple control surfaces through the dashboard, CLI, REST API and TypeScript/Python SDKs.¹²³ Teams can deploy ready-made Claude Code, OpenClaw, Hermes, ZeroClaw and DeepSeek Harness templates or ship their own Docker container; custom frameworks such as CrewAI or LangGraph can run through the generic container route.⁴²

The platform is built around fleets rather than one-off sandboxes. Agents keep persistent data across sleep, restart and redeploy, while triggers, channels and web requests can wake them as needed.⁵ Maritime also exposes persistent computer endpoints for model-driven desktops and file/action workflows.³ Vendor pages describe Firecracker micro-VM isolation and fast boot/wake behavior, but AiToolMap did not independently test isolation strength, latency or fleet reliability.¹

Maritime has a meaningful free tier: $0 for three agents with no card required. Starter is $20/month for 20 agents, Growth $100/month for 100 and Scale $500/month for 500; extra agents cost $1.50, $1.25 and $1.00 per agent/month respectively, while Enterprise is custom.⁶ Paid tiers can also add RAM, SSD capacity and always-on operation at published flat monthly rates.⁶

The pricing model is notable because it is capacity-based rather than metering tokens, messages or runtime seconds. That can make fleet budgeting predictable, especially when agents spend meaningful time asleep. The free three-agent allowance also provides a lower-friction evaluation path than many infrastructure products. Actual total cost still depends on memory, storage and always-on add-ons, so production buyers should model those separately.⁶

Maritime's Privacy Policy, last updated May 17, 2026, says the service collects account data, agent configurations, deployment metadata and logs, encrypted API keys/secrets, usage metrics and ordinary web telemetry.⁷ It names Stripe, Railway, AWS, Hetzner and Neon as service providers, describes TLS in transit and AES-256-GCM for sensitive credentials at rest, and states that agent deployment data and logs are retained for the life of the agent plus 30 days after deletion.⁷

The environment-variable documentation says secrets are encrypted at rest, injected only into the relevant agent VM and masked in the UI; secret values are blanked when packaging an agent as a reusable template.⁸ The Terms state that customers retain ownership of deployed code, data, configurations and content, granting Maritime only the limited license needed to host, store and transmit that content to provide the service.⁹

Those disclosures provide useful transparency but do not amount to an independent security audit. The Terms also expressly avoid guaranteeing uninterrupted, secure or error-free service.⁹ The fixed independent panel produced no qualifying exact-product evidence, which is treated neutrally under Rating v1.0. Y Combinator independently corroborates the active Fall 2026 company and the managed-agent-cloud positioning, but is not a performance or security benchmark.¹⁰

Maritime is best suited to developers building products where each customer or workflow needs its own long-lived agent, as well as platform teams managing larger fleets of isolated agent runtimes. The combination of SDK/API provisioning, lifecycle hooks, persistent storage and flat per-agent pricing is particularly relevant for multi-tenant agent products.²³⁶

It is less suitable for buyers that need independently certified infrastructure controls before deployment, teams that only need occasional ephemeral code execution, or organizations unwilling to rely on a young managed infrastructure provider. Sensitive deployments should evaluate the documented data-retention and subprocessor model and independently validate runtime isolation against their own threat model.⁷⁹