> ## Documentation Index > Fetch the complete documentation index at: https://aomi.dev/docs/llms.txt > Use this file to discover all available pages before exploring further. # Overview > Learn how plugins become Aomi Apps and how the runtime handles execution, signing, payments, and delivery across user surfaces. Out of the box, the harness turns language into verified onchain execution. That is powerful, but most real workflows need more than chain state. A trading agent needs an order book. A prediction-market agent needs the market's own API. A research agent needs price feeds, stats, news, and search. Your product's agent needs your product's backend. **Plugins extend the harness.** Like Claude connectors and the earlier ChatGPT plugin system, a plugin wraps your external APIs as tools the agent can call, with a system prompt that shapes how it uses them. The runtime loads plugins dynamically, next to the default agent, with no restart. Your tools run inside the same pipeline as everything else: every transaction they stage is simulated before anyone signs, and signing stays non-custodial. A deployed plugin combines your tools and prompt on the runtime. Aomi calls the deployed unit an **App**. Users select an App in the portal, CLI, and the Developer Platform. In these docs, "plugin" is the thing you write; "App" is the thing you ship. ## What a plugin contains A plugin is a small Rust crate compiled to a shared library. You declare it with one macro, `dyn_aomi_app!`, and one config file, `aomi.toml`. That is the whole contract. You write the tools; the platform handles the hard parts: * **Non-custodial wallets.** Your users keep their own keys. Your plugin never holds them. * **Simulate first.** Every transaction is simulated before anyone signs, so the agent shows the real outcome before it commits. ## Example: Somm Finance [Somm Finance](https://somm.finance) provides automated DeFi vaults that seek yield while managing risk. [Agentic Somm](https://agentic.somm.finance) helps you move idle USDC into Aave v3 on Base through a human-approved flow. Somm's models, data, and risk framework already existed as five HTTP endpoints. The team wrapped those endpoints as tools and used the investment mandate as enforced configuration in the system prompt. They deployed the plugin as the **Somm Liquidity Manager** App.