当我们在 2024 年启动 Archil 时，路线图非常简单：构建世界上最好的文件系统，与应用程序团队合作对其进行针对性优化，然后推出计算产品以提供一流的产品price/performance我们构建的工作负载略有不同（我们无法…
当我们在 2024 年启动 Archil 时，路线图非常简单：构建世界上最好的文件系统，与应用程序团队合作对其进行针对性优化，然后推出计算产品以提供一流的产品price/performance我们构建的工作负载略有不同（我们无法预料 bash 上的训练后模型会对文件系统产生难以想象的新需求），但路线图保持不变 我们真的很兴奋，真正的网络可寻址计算使我们能够简化用户在 Archil 之上构建更高级别基础设施的方式。
查看英文原文
when we started Archil back in 2024, the roadmap was super simple: build the world's best file system, partner with application teams to purpose-optimize for them, and then launch a compute product to deliver best-in-class price/performance the workloads we were building for were slightly different (we couldn't have anticipated post-training models on bash driving unimaginable new demand for file systems), but the roadmap has stayed the same we're really excited for how real, network addressable compute allows us to simplify how our users build higher-level infrastructure on top of Archil. persistent sandboxes will make it as simple as pressing a button to get your own: - agent hypervisor (running a "brain" loop w/ secure bash "hands") - postgres database - or even a "cursor origin"-like git remote it's deeply important to us that we're not only serving the best agent builders today, but that we're also building the substrate which will allow agents to build and manipulate stateful infrastructure themselves persistent sandboxes are the first step in that journey. please check them out, we're excited to see what you're able to build!
when we started Archil back in 2024, the roadmap was super simple: build the world's best file system, partner with application teams to purpose-optimize for them, and then launch a compute product to deliver best-in-class price/performance the workloads we were building for were slightly different (we couldn't have anticipated post-training models on bash driving unimaginable new demand for file systems), but the roadmap has stayed the same we're really excited for how real, network addressable compute allows us to simplify how our users build higher-level infrastructure on top of Archil. persistent sandboxes will make it as simple as pressing a button to get your own: - agent hypervisor (running a "brain" loop w/ secure bash "hands") - postgres database - or even a "cursor origin"-like git remote it's deeply important to us that we're not only serving the best agent builders today, but that we're also building the substrate which will allow agents to build and manipulate stateful infrastructure themselves persistent sandboxes are the first step in that journey. please check them out, we're excited to see what you're able to build!
when we started Archil back in 2024, the roadmap was super simple: build the world's best file system, partner with application teams to purpose-optimize for them, and then launch a compute product to deliver best-in-class price/performance the workloads we were building for were slightly different (we couldn't have anticipated post-training models on bash driving unimaginable new demand for file systems), but the roadmap has stayed the same we're really excited for how real, network addressable compute allows us to simplify how our users build higher-level infrastructure on top of Archil. persistent sandboxes will make it as simple as pressing a button to get your own: - agent hypervisor (running a "brain" loop w/ secure bash "hands") - postgres database - or even a "cursor origin"-like git remote it's deeply important to us that we're not only serving the best agent builders today, but that we're also building the substrate which will allow agents to build and manipulate stateful infrastructure themselves persistent sandboxes are the first step in that journey. please check them out, we're excited to see what you're able to build!
Today, we're excited to introduce Archil Persistent Sandboxes, making it simple to build full-stack agentic applications without having to leave the Archil cloud. They are the ideal way to run long-running code, like agent loops, alongside context stored in Archil disks.
- 来源
- X @ycombinator · x-ycombinator
- 证据角色
- x_project_network
- 时间状态
- known
- 用户判决
- 未判