> For the complete documentation index, see [llms.txt](https://strata-ai.gitbook.io/strata-ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://strata-ai.gitbook.io/strata-ai/high-level-guides/openapi.md).

# Environment

Strata AI introduces an enhanced **Environment** framework to support diverse applications of multiple agents across various scenarios, such as games, mobile platforms, and more. This extended functionality allows Strata AI to handle complex multi-agent interactions while remaining adaptable to unique operational requirements.

**Definition**

The **ExtEnv** (External Environment) is defined as the carrier for interactions between agents and external systems. For example:

* In a gaming scenario, ExtEnv connects the agent to a remote engine service and API interfaces, enabling direct interactions with the game.
* For role-playing games like Werewolf, ExtEnv can define game rules and manage states, such as tracking character statuses during different phases of the game.

The **XxxEnv** (Internal Environment) refers to the environment directly utilized by agents and teams. It typically inherits from ExtEnv and incorporates customizations for specific use cases, such as software development environments.

**Core Concepts**

* **Observation Space**: Represents the set of states obtained from the external environment.
* **Action Space**: Represents the collection of actions that can be performed in the environment.

By defining these spaces, Strata AI simplifies the abstraction of diverse environments, allowing developers to focus on implementing logic while users focus on leveraging space values.

**Features**

* **ExtEnv** integrates design principles from reinforcement learning frameworks like Gymnasium to streamline environment interactions.
* Decorators like `mark_as_readable` and `mark_as_writeable` ensure unified management of external environment interfaces, enabling agents to automatically call relevant APIs based on natural language inputs (upcoming feature).

***

####