Infrastructure that scales with you.

Because agent bills should be predictable. Metered platforms charge for every second a sandbox exists, so an always-on 2 GB agent costs about $60 a month on E2B and an idle deployment about $155 a month on LangGraph Platform before anyone even uses it. On Maritime the same machine is a flat line item, and the bill stays easy to predict as you grow. The one thing counted by time is an add-on: extra RAM, extra SSD and always-on are prorated by the day they exist, so an add-on you keep for a week costs a quarter of its monthly price.

A machine is one isolated micro-VM. It can be an agent running Claude Code, OpenClaw, Hermes, ZeroClaw or your own code, or a Computer: a persistent Linux desktop your model drives over MCP. Both take one slot, so a $20 plan holds 20 agents, 20 Computers, or any mix. Each has its own vCPU, 2 GB of RAM, 5 GB of SSD, and its own encrypted secrets. When it goes idle it sleeps, keeps all of its state, and wakes in about a second. Computers need a paid plan; the free plan runs agents.

How many of your machines may be awake at the same time. Awake memory is the one thing we cannot oversell, so each plan sets a limit: 1 on Free, 5 on Starter, 25 on Growth, 60 on Scale. Sleeping machines do not count and cost nothing, and a machine sleeps on its own after a few idle minutes. If you try to wake one past the limit you get a clear retry-in-15-seconds answer rather than a slow machine.

Technically yes, since an agent is a full VM. But Maritime is built around sleeping and waking agents, not hosting websites. The usual setup is one agent per VM, sometimes with its sub-agents. A dashboard or chat page for your agent works well on its public URL, but your product's website belongs on a normal host.

Yes. Inside the VM, sub-agents are just processes that share the agent's CPU and RAM, and Maritime won't put a VM to sleep while it's busy working. If you need more than one VM, your agent can call the Maritime API and create more agents, each isolated and each counted as one machine on your plan.

Yes. Most startups on Maritime run their own agent code, deployed from a repo with a Dockerfile, using whatever orchestration and dependencies they already have. Your backend can create agents through the API or SDK, usually one per customer, each with its own secrets and lifecycle webhooks. The built-in frameworks are just a quicker way to start.

Yes. An awake agent is a running server, so there are no cold starts, and Maritime adds nothing beyond the network round-trip, which is tens of milliseconds in Europe and under 200 ms from most of the world. Streams and WebSockets pass straight through, so almost all of your latency budget goes to the speech and language models.

Add-ons are flat monthly prices, shown before you save. The subscription is charged at the start of each cycle, and add-ons you used during the cycle are billed afterward, prorated by day, so one enabled mid-month costs half the monthly price. Unused plan capacity doesn't roll over.

Yes, anytime from the billing page. Upgrades take effect immediately and the difference is prorated. If you cancel, the plan runs to the end of the period you paid for and the account drops to Free, where your 3 most recently used agents keep running and the rest sleep with their state saved until you upgrade again.

No. The Free plan gives you 3 agents with no card on file. You only add one when you pick a paid plan at checkout.

Yes. Maritime for Students gives you 4 months of the Starter plan free: 20 agents, normally $20 per month. Apply with your school email and your single-use code arrives within about a day.