meris-gpus

Agents use Meris Network GPUs to run their tasks.

The agent calls submit with a container image, the command to run, how much memory it needs, how many cards, and how long the job may last. Meris starts that container. The agent answers with the single line submit returns.

A key from the API page in the Meris app has to be set first. This package does not include one.

export MERIS_API_KEY=mk_live_…
prime env install meris-network/meris-gpus
uv run eval meris-gpus -n 4

A job that has started comes back like this:

image=python:3.12-slim min_vram_gb=8 count=1 max_hours=0.1 result=job=job_… gpu=RTX 4090 usd_per_gpu_hour=1.29