Introducing Persistent Sandboxes. Archil introduces traditional, forkable sandboxes with persistent disk and ability to host network services. Today, Archil is releasing Persistent Sandboxes. Persistent Sandboxes are the easiest way to execute long-running untrusted model-generated code, network services, or interactive development environments that connect to data on your existing Archil disks. Archil already offers Serverless Execution for customers who want to give their agents a run_bash that works on real Linux machines without needing to manage a sandbox. However, after working closely with some of the best teams building agentic products, Archil has identified that some workloads don't actually fit a one-off execution tool. For example, teams need a place to actually run their agent loop, start development servers with preview URLs, or keep expensive-to-start processes alive in memory. Persistent Sandboxes fill this gap with a familiar API for existing sandbox users, including the ability to specify an OCI base image, vCPU count, and memory. The sandbox remains available across calls, so that long-running processes remain running while the agent is working. If your agent stops or needs to pause, you can stop the sandbox and stop paying for compute without losing its files or installed dependencies. When used with an agent loop, Persistent Sandboxes stay alive in between turns, which means that it can follow the entire trajectory of long-running agents, but they can also be paused with memory snapshotting or forked so that you can preserve known good states or explore multiple approaches in parallel without rebuilding the environment each time. If your agent produces something interactive, you can also generate preview URLs inside of the sandbox so that you can access network services from the public internet. For more complex tasks, Persistent Sandboxes can be combined with Archil disks and Serverless Execution. Teams often store agent context on an Archil disk and need that data to remain strongly consistent across several environments such as: * Humans accessing a website that uses the Archil S3 API * The agent's bash tool operates on the context with Archil Serverless Execution * The agent loop can now run and mount the same disk inside a Persistent Sandbox Many of its customers are currently choosing to use Archil with an existing sandbox platform like E2B or Daytona, and this launch only strengthens its commitment to deliver Archil as an open, neutral storage platform for all sandbox providers. Archil continue to recommend using other sandbox providers for tasks that require hundreds of thousands of concurrent vCPUs, Windows or macOS tasks, or attached GPUs, but Archil is excited to deliver a simple primitive for Archil customers who "just" need a Linux box. Persistent Sandboxes are a powerful new way to run agents entirely inside of the Archil cloud, and they're available immediately for Archil customers in AWS regions. Get started now by generating an API key on the Archil console, or contacting its sales team to get BYOC set up in your cloud in about an hour.

Archil has raised $11 million in Series A funding led by Standard Capital, with participation from YCombinator, Felicis, Peak XV Partners and Wayfinder Ventures. The round brings total capital raised to $18 million, less than a year after its seed funding. The startup is developing file system infrastructure designed specifically for AI applications, including inference, model training and agents. Archil claims file systems offer the best way for AI agents to interact with data, as models are inherently trained to work with files and folders. The company has launched serverless execution, allowing users to treat file systems as a service that accepts bash commands and returns results. Archil's platform synchronises with Amazon S3 and is designed to connect AI systems to large datasets.