Coding agents
Give every Claude Code or Codex task its own machine. Fork a prepared VM into fifty copies and run them side by side.
flock prompt --fork base -n 50 "fix the failing tests"Early access EU-hostedA flock of VMs.
A flock of VMs.
For your agents.
Real Linux microVMs that boot in under a second. Run coding agents, CI jobs and batch work in parallel, and pay by the second only while they fly.
Free to join. Early users get free hours.
Flock· fr-par · now
$ flock new --from node-22 --fork 30.6 s
Running
38
Landed today
1,204
Spend today
$4.12
- fv_8k2mclaude · billing refactorrunning
- fv_8k2nclaude · fork 2 of 3running
- fv_8k2pci · e2e tests #4512running
- fv_7x1qcodex · docs passlanded
< 1 s
from request to SSH, out of a warm pool
$0.016
per hour for 2 vCPU, billed by the second
14 days
of snapshots kept after a VM stops
6 regions
Paris, Amsterdam, Warsaw, Virginia, Chicago, San Jose
Use cases
One machine. Four ways to fly it.
CI runners
Change one line in your GitHub workflow. Each job gets a fresh VM that is deleted when the job ends.
runs-on: flockAlways-on services
Bots, workers and preview servers with a public HTTPS URL. Keep them up with systemd, like any Linux box.
flock new --no-auto-stop --expose 3000Off-peak batch
Evals, RL rollouts and big builds that can wait. They start within 24 hours, at half price.
flock batch run.sh --count 400How it works
Three steps. No servers to babysit.
- 01
Ask for a VM
From the CLI, the API or a GitHub workflow. Pick a size, a snapshot to start from and how long it may idle.
- 02
Let it work
Your agent or job gets root on a real Linux machine with Docker, systemd and its own IPv6.
- 03
It lands
Idle VMs stop on their own and keep their disk. Billing stops with them, to the second.
A real machine
Its own kernel, root, Docker, systemd and /dev/kvm. Not a container that breaks on the first apt install.
Snapshots and forks
The disk is saved while the VM runs. Stop it, resume it tomorrow, or fork it into copies.
Keys stay outside
Connect your Claude or ChatGPT plan once. A proxy adds the real key, so the VM only holds a dummy token.
Isolated by hardware
Each VM runs in its own Firecracker microVM. A compromised agent cannot reach the host or other customers.
Public URLs
Expose any port on a stable HTTPS address, with a token if you want one.
Webhooks and OTel
Get an event when a VM is ready, stops or fails, and send its metrics to your own stack.
Pricing
Pay for seconds. Not for servers.
The plan is $20 a month and every dollar comes back as VM time. Egress, IPv6 and 14 days of snapshots are included. Off-peak batch runs at half price.
| Size | vCPU | RAM | Disk | Per hour | On 24/7, per month |
|---|---|---|---|---|---|
| small | 2 | 4 GB | 50 GB | $0.016 | $11.68 |
| default | 4 | 8 GB | 50 GB | $0.032 | $23.36 |
| large | 8 | 16 GB | 110 GB | $0.064 | $46.72 |
| xlarge | 16 | 32 GB | 250 GB | $0.180 | $131.40 |
A 6-minute CI job on small costs $0.0016. A 2-hour agent session on default costs $0.064.
FAQ
Questions. Answered.
How is this different from a container sandbox?
Each VM has its own kernel and hardware isolation through KVM. Agents can run Docker, systemd or even nested VMs, and a compromised VM cannot reach the host or other customers.
Can I use my Claude Max or ChatGPT subscription?
Yes. Run claude setup-token on your machine and paste the token once. You can also bring an Anthropic or OpenAI API key.
What happens when a VM is idle?
Agent VMs stop after the idle time you choose, 30 minutes by default. The disk is kept for 14 days so you can resume where you left off.
When can I start?
We are onboarding early users in small groups. Join the list and we will send your access with free hours to try it.
Let your agents fly.
We open access in small groups. Early users get free hours and a direct line to the team.