Backed by Signal Ventures
Persistent machines for autonomous software.
Give every agent, workflow and user its own always‑on VM. It boots in half a second, remembers everything, and can’t break anything that matters.
kernal — zsh — 80×24
Trusted by teams shipping agents in production
Autonomy
Give your agents true autonomy. Their own space to explore, build, break things, and try again.
Persistent workspaces. Files, repos, browser profiles and auth that stays put across sessions. An agent can come back after a week and pick up where it left off.
Download and run software. apt‑get any library you need, run headless Chromium, clone repos, compile and ship. It’s a real Linux box, not a sandboxed function.
Run AI‑generated code safely. Let your agent write and execute its own tooling, serve web apps and APIs, or run scripts it wrote itself, without touching your infrastructure.
Actual isolation. Every machine is a hardware‑isolated microVM. Give the agent autonomy to try anything and safely clean up along the way.
Machines
Every agent gets a real, persistent machine that’s up in half a second.
An actual VM.
A Firecracker microVM with its own kernel, exclusive vCPUs, private network and preinstalled dependencies.
vcpu
2 – 32
memory
1 – 128 GB
guest kernel
6.8 · hardened
network
private · egress‑filtered
state
running
A persistent disk.
Everything written to /workspace survives sleeps, restarts and deploys. Snapshot it, fork it, roll it back.
$ du -sh /workspace/*
memory/ 1.2G
repos/ 4.8G
browser/ 812M
packages/ 2.1G
snapshots/ 14
40 GB · persistent · encrypted
Really fast boot and wake.
Machines sleep when idle and wake on the first request. No cold containers, no warm pools to babysit.
0ms
cold boot
0ms
wake from sleep
kernal
0.4 s
container cold start
6 – 12 s
classic vm
45 – 90 s
Interfaces
Run your agent here. Dashboard, CLI, and SDK.
Dashboard
Deploy from a template, connect a repo, watch logs and files live.
CLI
One command from a template or repo, then tail it live in the shell.
TypeScript SDK
Give each user their own machine in a few lines. Streams, files and exec included.
Python SDK
The same API from Python, with async support for agent frameworks.
machines · us‑east‑1
12 running
name
status
agent‑research‑01
running
browser‑worker‑07
running
nightly‑evals
running
support‑bot‑eu
sleeping
scraper‑pool‑3
running
fine‑tune‑runner
sleeping
MCP
Or keep your agent, but give it a computer. One MCP server. A desktop per user id.
claude_desktop_config.json
Setup
Add one MCP server to the agent you already have. No SDK, no migration.
In use
The agent gets a computer: browser, shell, files. Per user, per session, isolated.
Hand over
Your user can open the same desktop, see what happened, and take the wheel.
Give your agents a dedicated desktop.
desktop.kernal.run/u/user_8f21
AGENT LOG
10:41:02 opened booking.example.com
10:41:09 filled form #checkout (4 fields)
10:41:15 screenshot saved /workspace/run‑117.png
10:41:16 waiting for human → hand over
