An agent picks up a task it started last week. It needs its files, the software it installed and the browser it’s still signed into. [1]
For a startup selling agents, every customer needs one of those computers. Setting them up, keeping them running and debugging each one eats the team’s time. [2]
Maritime gives every agent its own computer.
It’s a fully managed cloud where each agent runs in its own isolated, persistent virtual machine, from $1 an agent a month. Maritime handles the infrastructure, scaling, sleep, wake and monitoring. [2][3]
Instead of rebuilding an environment for every task, the agent comes back to its own machine. [1]
Its own machine, kept between tasks
Each agent gets an isolated VM where it can run code, install software and use a browser. Its files, memory, installed software and logins stay between tasks. [2]
Teams choose the machine the work needs: a Linux computer, a Windows desktop or a virtual Android phone. [2]
The agent can run inside the VM, or stay on the team’s own backend and control a Maritime computer through MCP. A person can take over the screen to finish a step, with the agent’s actions paused until they’re done. [2][5]
Idle agents sleep and wake when work arrives, so a parked agent costs only its storage. Agents that must run around the clock can stay always on. [1][4]
Teams can inspect, debug and fix any one customer’s agent directly, and the first three agents host for free. [2]
Built by researchers who hosted hundreds of agents at MIT
Maria Gorskikh was a visiting lecturer at MIT and a researcher at its NANDA lab, where Aryaman Shrivastava researched AI agent coordination at scale. [3]
They started Maritime when OpenClaw came out. During their research they had hosted hundreds of agents, and even at that scale their time went into getting environments running, keeping them up and debugging individual agents. [2]
They still publish. A recent paper by the founders shows how snapshotting idle agents to disk lets a parked agent cost only its storage; a typical agent in their fleet is awake about eleven minutes a day. [1]
Maritime is based in San Francisco and is part of Y Combinator’s Fall 2026 batch. [3]
What this looks like in practice
A Windows computer for every dental office
A startup sells an AI agent to dental offices. Each office’s agent gets its own Windows computer and works in the practice’s existing desktop software, the way a person would. [6]
The agent returns to the same computer, files and tools for that office on every task, and a staff member can take over the screen when a step needs a human. [5][6]
A personal assistant that remembers
A team building a personal assistant gives each user their own VM. Asked to move a 9 a.m. meeting, the assistant moves it to 10, with the user’s calendar and email still signed in and a journal of what it has done ready for next time. [1]
Where Maritime ranks
Most sandboxes spin up for a task and disappear. Maritime gives each agent a computer that stays. [1][2]
The bigger picture
Everyone will have agents working for them, and businesses will run whole teams of AI employees, Maritime’s founders believe. [2]
Each of those agents needs a computer of its own, and someone has to run millions of them. [2]
Maritime is betting it can make that as easy as launching the agent itself, for about a dollar a month per agent.
One agent, one computer, for as long as it needs it.
Written by SOTA2 from Maritime’s public materials and information the company provided. Customer experiences and product plans are reported by Maritime. Ranking analysis and comparison data are from SOTA2.
Sources
- Maritime — host persistent agents, each in its own VM
- Launch YC — Maritime: fully managed cloud for running AI agents at scale (October 2026)
- Y Combinator — Maritime company and founders
- Maritime documentation — agent lifecycle and persistence
- Maritime documentation — Computers, identity and takeover
- Maritime — product description and dental-office scenario provided for this article by the company
