# inference shell

> the ai runtime that compounds with every session. run any model, compose agents, stack knowledge - it never forgets.

inference shell is a platform for building and running ai agents in production. pre-built apps (image, video, audio, text, search, 3D), durable execution, human-in-the-loop approval, real-time streaming, and a skill registry that improves with use.

## belt cli - the fastest way to use inference shell

if you're an agent or developer, belt is the easiest way to interact with inference shell.

install: `curl -fsSL https://cli.inference.sh | sh`

```bash
# run any ai model
belt app run pruna/flux-dev -i '{"prompt": "a cat"}'

# search apps
belt app search "image generation"

# use skills (versioned instructions that improve with use)
belt skill use my-org/my-skill

# connect MCP servers
belt mcp add github
```

learn more: https://inference.sh/belt

## what's on the platform

- **apps** (https://inference.sh/apps): any ai model. one api call. image, video, audio, text, search, 3D. one key, pay per run, zero vendor lock-in.
- **skills** (https://inference.sh/skills): don't write the same prompt twice. proven instructions that just work. versioned, security-scanned, fitness-ranked.
- **agents** (https://inference.sh/agents): from demo to production in an afternoon. durable execution, persistent state, human-in-the-loop.
- **ui** (https://inference.sh/ui): ai-native react components. chat, generative ui, tool approvals. 30+ widgets.
- **commons** (https://inference.sh/commons): your team's shared ai workspace. shared memory, automations, self-hostable.
- **belt cli** (https://inference.sh/belt): one install. your agent never starts cold again. skills, knowledge, and apps surface automatically.
- **MCPs** (https://inference.sh/apps): MCP servers, OAuth integrations, one-command setup.

## navigate

- docs: https://inference.sh/docs (start here: https://inference.sh/docs/getting-started/introduction)
- blog: https://inference.sh/blog
- apps: https://inference.sh/apps (by category: https://inference.sh/apps/category/<category>, by task: https://inference.sh/apps/tag/<task>)
- gpu pricing: https://inference.sh/gpus (weekly gpu price index: https://inference.sh/gpus/price-index)
- pricing: https://inference.sh/pricing (markdown: https://inference.sh/pricing.md)
- api reference: https://inference.sh/docs/api/authentication

## links

- website: https://inference.sh
- github: https://github.com/inference-sh
- discord: https://discord.gg/inference
- python sdk: `pip install inferencesh`
- js sdk: `npm install @inferencesh/sdk`

## markdown

every page below has a markdown version at its url plus `.md`, and answers `Accept: text/markdown` with it.

- home: https://inference.sh/index.md
- pricing, with every app's price: https://inference.sh/pricing.md
- apps: https://inference.sh/apps.md (categories: https://inference.sh/apps/category/<category>.md, tasks: https://inference.sh/apps/tag/<task>.md, makers: https://inference.sh/apps/<maker>.md, one app: https://inference.sh/apps/<maker>/<app>.md)
- docs: https://inference.sh/docs/index.html.md (each page: https://inference.sh/docs/<path>.md)
- blog: https://inference.sh/blog/index.html.md (each post: https://inference.sh/blog/<path>.md)

## optional

- [llms-full.txt](https://inference.sh/llms-full.txt): complete documentation and blog content for deeper context