CURATED / DEPLOYMENT LAB

Local first.
Cloud when it makes sense.

Understand the workflow before paying for hardware. These guides and our model directory are free to read.

COMING SOON

Runpod integration

One place for reviewed deployment recipes, memory planning and reproducible test conditions. Automated provisioning is not available.

Runpod compute and storage are paid.

We have not created a Pod, opened a paid account, entered payment details or rented a GPU. Nothing on this page starts a billable service.

SOURCE REVIEWED / NOT CLOUD TESTED

llama.cpp

Dense GGUF models

Start locally, size context and compare an equivalent request before considering rented hardware.

Ollama

Supported text and vision models

Check the model license, local API and selected quant. The installed model determines the memory requirement.

ComfyUI

Supported image/video pipelines

Review workflow nodes and checkpoints before running them. Third-party nodes execute code and require their own review.

Choose the exact checkpoint, quantization and license.
Measure a local prompt, generation speed and total task time.
Calculate weights, KV cache and runtime overhead, leaving headroom.
Keep inference endpoints private or authenticated; never expose an unprotected model API.
For future cloud experiments, review compute, idle storage and shutdown rules before authorizing any spend.