tipod

Every agent will need its own computer.

tipod gives each AI agent a full Linux pod — a real computer with its own disk, saved every minute, billed per second while running, and back online in seconds with everything intact.

$ curl -fsSL https://tipod-control-plane.ramasamy-vignesh.workers.dev/install | sh

Then pod create and pod shell active. You are root. Read the quickstart.

The product

A real computer per agent

Each pod is a KVM virtual machine — not a container and not a sandbox that evaporates. SSH in as root with apt, systemd, and Docker on Ubuntu 24.04. The disk is saved about once a minute, so a stopped agent is a paused thought, never a lost one.

isolation

One VM per agent. A hypervisor boundary, not a permission flag — a compromised agent stays inside its own pod.

abort-stop

A pause that cannot complete its final save never succeeds. The pod reverts to running and records pod.stop_failed instead of losing the work.

machine-time

Per-second billing while running. Paused pods cost nothing, and unused seconds carry forward.

Lifecycle

Eight commands, start to finish

active refers to the last pod you created in this shell, so you rarely type an ID.

$ pod create --type small --name api-worker --ttl 900
✓ Created pod tp_abc123 · type small · state creating

$ pod shell active
Welcome to Ubuntu 24.04 LTS — you are root on your own machine.

$ pod expose active 3000 --title "Agent preview"
✓ https://host-tp_a1b2c3.tipod.dev?_token=…

$ pod pause active
✓ Disk saved · state archived · billing paused

$ pod wake active
✓ Pod ready — same disk, same files

$ pod clone active --name api-worker-2
✓ Forked from the saved disk → tp_def456

$ pod quota
balance 1,842,300 s · active pods 2 / 200

$ pod rm active

Also available: pod ls, get, ttl, log, cp, tunnel, snap ls, tpl, and key. See the full CLI reference.

Pricing

Pay for run time. Keep the disk when you pause.

From $0.018/hour

A $20/month plan credits 2,000,000 seconds of machine time to your account. Pods spend it only while running — paused pods cost nothing, and unused balance carries forward.

Machine-time rates and balance behaviour
small — 2 vCPU / 4 GB$0.018 / hour
default — 4 vCPU / 8 GB$0.036 / hour
paused pod$0 — disk kept
unused balancecarries forward, never resets

Friends beta. Signup is invite-gated and grants trial machine-time. Ask the operator for a token, then run pod signup --email you@example.com --invite <token>.

Open the dashboard Read the docs

FAQ

Why does every agent need its own computer?

Shared hosts couple agents together. A compromised or runaway process can reach another agent's files, secrets, and network. One VM per agent keeps failures and credentials inside a hard boundary — and gives each agent a stable disk to grow into.

How is this different from a container sandbox?

Each pod is a full KVM/QEMU virtual machine with its own disk and systemd. You get machine semantics — a memory that persists across sessions — not a short-lived process tree that vanishes when the orchestrator restarts.

What is abort-stop?

pod pause is asynchronous: the pod moves to archiving while a final save is written, then lands in archived. If the save or shutdown fails, the pause aborts — the pod returns to running and billing continues. A pod is only ever archived when the save is verified.

Am I billed while a pod is paused?

No. Paused pods cost nothing. You pay per second only while the machine is running, and pod wake restores the same disk in seconds.

What survives pause and wake?

Files, installed packages, configuration, and enabled systemd services. Saves are disk-level, so processes you started by hand need restarting — and RAM state, hostname, and SSH host keys are not preserved.

Where do pods run?

APAC-first, with Singapore staging today. Dedicated public IPs are on the roadmap; for now SSH uses the host IP with a mapped port.