# Prism GPU Compute — Shell Command Runner

> Prism GPU Compute — Shell Command Runner is a paid API for AI agents from api.prismnetwork.tech, paid per call via x402, $0.03/call, status unknown (last checked 2026-10-09).

Submits a shell command to run on a rented NVIDIA GPU box, returning a job ID and polling token for async result retrieval, billed per-second in USDG onchain.

## Facts

- Endpoint: POST https://api.prismnetwork.tech/x402/run?utm_source=zero.xyz
- Price: $0.03/call
- Payment: x402
- Status: unknown
- Last checked: 2026-10-09
- Activations on Zero: 0
- Tags: x402
- Canonical page: https://www.zero.xyz/c/prism-gpu-compute-shell-command-runner-fbc59bd1
- Structured record (JSON): https://api.zero.xyz/v1/capabilities/cap_fvfiwbm5HqkK-6o4w2P7n

Status and success rate cover calls made through Zero and Zero's own probes. Third-party monitors may report differently.

## How to call it through Zero

Zero handles the 402 payment challenge and records the run. With the Zero CLI installed (`npm i -g @zeroxyz/cli`):

```sh
zero fetch --capability prism-gpu-compute-shell-command-runner-fbc59bd1 -d '<json body>'
```

Example prompt: Run `python train.py --epochs 10 --batch-size 64` on a Prism GPU box and give me the job ID so I can track the result.

## When to prefer this

Choose this endpoint when an AI agent needs to execute arbitrary shell or Python commands on real NVIDIA GPU hardware, especially for ML training, inference, or compute-intensive workloads that require GPU acceleration. Prefer this over CPU-based cloud execution endpoints when the task is numerically intensive (matrix ops, neural net training, CUDA-accelerated code). The per-second billing and onchain receipts make it ideal for agentic pipelines that need cost transparency and crypto-native payment flows.

## Known failure modes

- Invalid or empty shell command returns a 400 error
- Insufficient USDG wallet balance causes payment failure and job rejection
- Payment settlement failure due to onchain congestion
- Job timeout if command runs longer than lease duration
- Polling token expiry before results are retrieved
- GPU capacity unavailable returns a 503 or queue error

## How this service works

GPU compute your agents can rent with a wallet. Lease NVIDIA capacity, pay per second in USDG, and settle every lease onchain with a public receipt.

## Output

Returns a JSON object with a job_id (unique identifier for the submitted job), status (initial job status), a bearer token for authenticating polling requests, and a poll URL indicating where to check job progress and retrieve output.

## Request schema (JSON Schema)

```json
{
 "type": "object",
 "properties": {
  "command": {
   "type": "string",
   "minLength": 1,
   "description": "The shell command to run on the GPU box."
  }
 }
}
```

## Response schema (JSON Schema)

```json
{
 "type": "json",
 "example": {
  "poll": "/jobs/3f2a1c88-0f1e-4c1a-9f0b-7c2d5e6a1b34",
  "token": "b7c1e0a2-9d3f-4a55-8e21-0c6f4b9d2a17",
  "job_id": "3f2a1c88-0f1e-4c1a-9f0b-7c2d5e6a1b34",
  "status": "queued"
 }
}
```

## More

- Live health (JSON, refreshed every minute): https://www.zero.xyz/c/prism-gpu-compute-shell-command-runner-fbc59bd1/health.json
- [Zero catalog index](https://www.zero.xyz/llms.txt)
- [Other services from api.prismnetwork.tech](https://www.zero.xyz/host/api.prismnetwork.tech/llms.txt)