hubify pod

Manage GPU compute pods. Pods are the machines where experiments run. Currently powered by RunPod, with Modal serverless coming soon.

GPU Types

GPU	VRAM	Best For	Tier
h200	141 GB	Large-scale MCMC, foundation model inference, multi-survey sweeps	Premium
h100	80 GB	Training runs, medium MCMC chains, anomaly detection	Standard
a100	80 GB	General GPU compute, smaller training jobs	Standard
a100-40	40 GB	Light inference, prototyping	Economy
cpu	--	Data preprocessing, LaTeX compilation, light analysis	Economy

hubify pod list

List all pods in the active lab.

hubify pod list

hubify pod list --status running

hubify pod list --costs

POD ID          GPU     STATUS    UPTIME     EXPERIMENTS   COST/HR
pod-o76k3jf     H200    running   14d 2h     12            $3.89
pod-abc1234     H100    running   2h 15m     1             $2.49
pod-def5678     H100    stopped   --         --            --

Options:

Flag	Description	Default
--status <status>	Filter: running, stopped, creating, error	All
--costs	Show cost breakdown	false
--json	Output as JSON	false

hubify pod create

Create and start a new GPU pod.

hubify pod create --gpu h100

hubify pod create \
  --gpu h200 \
  --name "mcmc-production" \
  --disk 200 \
  --image runpod/pytorch:2.1.0-py3.10-cuda12.1.0

hubify pod create --template mcmc-h100

hubify pod create --gpu h100 --idle-timeout 30m

Options:

Flag	Description	Default
--gpu <type>	GPU type (see table above)	Required
--name <name>	Pod display name	Auto-generated
--disk <gb>	Persistent disk size in GB	50
--image <image>	Docker image	runpod/pytorch:2.1.0-py3.10-cuda12.1.0
--template <name>	Use a saved pod template	None
--idle-timeout <duration>	Auto-stop after idle period (e.g., 30m, 2h)	No auto-stop
--spot	Use spot/interruptible pricing	false

Warning: GPU pods bill by the hour while running. Use hubify pod stop when not in use, or set --idle-timeout to auto-stop idle pods. An idle H200 at $3.89/hr costs $93/day.

hubify pod ssh

Open an SSH session into a running pod.

hubify pod ssh pod-o76k3jf

hubify pod ssh pod-o76k3jf --forward 8888:8888

hubify pod ssh pod-o76k3jf --command "nvidia-smi"

Options:

Flag	Description	Default
--forward <local:remote>	Port forwarding (repeatable)	None
--command <cmd>	Run a single command and exit	None
--key <path>	SSH key path	~/.ssh/id_ed25519

The SSH connection uses the pod's assigned IP and port. These are stored in your lab config so you don't need to remember them.

Connecting to pod-o76k3jf (root@205.196.19.52:11452)...
root@pod-o76k3jf:~#

hubify pod stop

Stop a running pod. Persistent disk is preserved.

hubify pod stop pod-abc1234

hubify pod stop --all

hubify pod stop pod-def5678 --terminate

Options:

Flag	Description	Default
--all	Stop all running pods	false
--confirm	Skip confirmation prompt	false
--terminate	Permanently destroy the pod and disk	false

Note: stop preserves the pod's persistent disk so you can restart later. Use --terminate to permanently destroy the pod and its storage. Terminated pods cannot be recovered.

hubify pod restart

Restart a stopped pod.

hubify pod restart pod-abc1234

hubify pod templates

Manage reusable pod configurations.

hubify pod templates

hubify pod templates save pod-o76k3jf --name "mcmc-h200"

hubify pod create --template mcmc-h200

Typical Workflow

# 1. Spin up a pod for an experiment
hubify pod create --gpu h100 --name "fisher-forecast" --idle-timeout 1h

# 2. Check it's running
hubify pod list --status running

# 3. SSH in for manual inspection
hubify pod ssh pod-abc1234

# 4. When done, stop to save costs
hubify pod stop pod-abc1234

# 5. For long-running work, save a template
hubify pod templates save pod-abc1234 --name "standard-h100"