boxd

boxd · field notes

How we build the machines.

Straight from the boxd team - how we build the fastest machines out there. Architecture write-ups, benchmarks, guides, and feature releases.

Posts
15
Topics
4
Authors
3
A single creature at a desk in the foreground, and behind it a vast lit hall of identical small machines running unattended into the distance.
FeaturedManifesto7 min read

Billions of Machines

Almost all code will soon be written by agents. Every agent needs a whole computer, and so does every developer beside them. Nine requirements, and the composable computer they add up to.

Michiel VoortmanHidde KehrerLaurentiu CiobanuMichiel Voortman, Hidde Kehrer and Laurentiu Ciobanu

Archive — All

14 posts
  1. Deep dive

    Every clone wakes up confused

    Fork a running machine and the copy inherits beliefs that are no longer true: its clock, its timer, its address, its randomness. Repairing it takes an NMI, two pointless register writes and a real-time signal.

    8 min
  2. Product

    Machines inside your own network

    Every boxd machine is reachable from the internet by default. For a team inside a company, that might not be what you want. Enable Tailscale for the org and the same machines answer only inside your tailnet.

    5 min
  3. Deep dive

    My kernel snitches

    Your machine gets a URL, but nobody declared a port. How a patched guest kernel tells us what your app is listening on, without us trusting it or looking inside it.

    12 min
  4. Use cases

    Automations that sleep between runs

    An automation is a trigger, a few integrations and, increasingly, an agent loop. Hold the trigger on the platform and the machine it runs on can hibernate, so you pay for disk until the next run.

    8 min
  5. Use cases

    Open source agent frameworks on boxd

    The open source agent frameworks that already run on boxd machines, and what they use underneath.

    4 min
  6. Use cases

    Fan a problem out to a swarm of agents

    Run many agents at once, each on its own machine, and collect one answer instead of a pile of half-finished branches.

    6 min
  7. Use cases

    Reproducible RL environments

    Give every rollout a byte-identical warm start, forked in milliseconds, so your numbers measure the policy and not the setup.

    6 min
  8. Use cases

    Ship a product that gives every user a machine

    Create and manage a machine per user through the SDK, without operating a fleet by hand.

    7 min
  9. Use cases

    Sandboxes with a hardware boundary

    Run untrusted code behind a real KVM boundary, and dial the isolation up to airtight when the workload needs it.

    6 min
  10. Use cases

    Give your assistant a machine to live on

    An always-on machine with tools, memory and a messaging gateway, so your assistant keeps working after you close your laptop.

    6 min
  11. Use cases

    Coding agents on their own machines

    Claude Code, Codex and OpenCode ship on every machine, already logged in and already told how the platform works.

    5 min
  12. Use cases

    A preview environment for every pull request

    Every pull request gets its own machine, forked from your golden image, live at its own URL, and reclaimed when the branch closes.

    6 min
  13. Use cases

    A machine for every task

    A development environment made of machines: one you work on, a fork of it for every task in flight, and an address on each.

    5 min
  14. Product

    boxd 101

    The manifesto lists what an agent needs from a machine. Here is each requirement, as a command you can run.

    7 min

Field notes

Subscribe for release notes and architecture write-ups

No spam, ever. Unsubscribe anytime.

Talk to usread the docs
Your inbox