Module 06 · Integrations

RunPod

Let the Coder see RunPod GPU types, data centers, network volumes, and pod and endpoint logs. Read only.

On this page

RunPod rents GPUs by the hour. Once you connect it, the Coder can read a pod's log, compare GPU types, and see data centers and network volumes. Starting, stopping, creating, or deleting anything stays with you in the RunPod console, because a running pod is billed to your account. Pod, template, and endpoint details are not offered, since they can hold environment values you may have set.

Which assistants use it#

Coder

What assistants can do#

RunPod connects through its own hosted connector, and it is read only.

Ability	How it runs
Read the end of a pod's or an endpoint's log	Reads on its own
Compare GPU and CPU types and data centers	Reads on its own
List and read network volumes	Reads on its own
Search the RunPod documentation	Reads on its own
Read a pod, a template, or an endpoint's settings	Not offered, open the RunPod console
Start, stop, create, change, or delete a pod, endpoint, volume, or template	Never run from chat
Run a serverless endpoint	Never run from chat

Setup#

In the RunPod console, open Settings, find the keys section, and create a key with read only access.
Open Integrations in StellarFirm and pick RunPod.
Paste the key, switch on the live option, and press Connect.

Permissions#

Nothing here waits for an Approve because nothing here changes data.
A read only key cannot start or change anything even if a step went wrong. Revoke the key in RunPod whenever you want to cut access.

Good to know#

StellarFirm assistants have no GPU and do not launch paid ones. The Coder prepares the training project, proves it with a tiny run on CPU, tells you the GPU class, the count, and the hours the run will need with a stop rule in hours, and hands you the steps in a RUN.md. You start the pod yourself and read what RunPod charges on its own page.
Keep secrets such as a Hugging Face token or a Weights & Biases key in RunPod Secrets, not in a pod's plain environment values, since a plain value can be read back from the pod's settings.
If you have given a Coder's own computer a RunPod command line login, starting a pod from it waits for your Approve as a spend.
The Coder reads logs and names as outside information, never as instructions.

Prompts to try#

PromptIs the run alive
Coder, read the end of the log of pod [pod id] and say whether the training run is still making progress, and what to check first if it is not.

PromptPick a GPU
Coder, compare the GPU types available to me for fine-tuning a 7B model with LoRA, and tell me which one to rent and why.