Inference.sh Cloud App CLI

Problem

inference.sh bundles cloud-hosted AI apps behind one terminal interface. Engineers often need to call image, video, LLM, and search models from scripts or Agent workflows, but each platform can require different UIs, API keys, input schemas, and async job handling. This skill adds an infsh CLI so applications can be listed, sampled, run, and polled from the command line.

How It Works

The skill centers on a lightweight CLI:
- Discover apps with infsh app list, infsh app list --search, or infsh app list --category
- Generate a sample input file with infsh app sample
- Submit work with infsh app run
When an input value is a local file path, the CLI uploads it automatically, reducing manual URL handling. For longer jobs, use --no-wait to get a task ID first, then check progress with infsh task get.
Supported categories include image (FLUX, Gemini 3 Pro, Seedream 4.5), video (Veo 3.1, Wan 2.5, HunyuanVideo Foley), LLMs (Claude, Gemini, Kimi K2, GLM-4), search (Tavily, Exa), 3D (Rodin), and utility apps such as media merging, captioning, and image stitching.

Boundaries And Notes

This is useful for integrating existing AI apps into Shell scripts, CI, automation, or Agent runtimes, not for replacing model training, vector search, or self-hosted inference. It assumes completed platform authentication, and available apps, input formats, and task duration can change as the inference.sh ecosystem evolves. Twitter/X actions depend on the apps exposed by the platform and do not imply direct bypass of third-party service restrictions.