> For the complete documentation index, see [llms.txt](https://help.globalnaturewatch.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.globalnaturewatch.org/get-started/models.md).

# Models

Global Nature Watch (GNW) Horizon is powered by a multi-model AI architecture designed to use the right model for each task—rather than relying on a single system for everything.

Instead of treating models as interchangeable alternatives, we run multiple specialized models in parallel, each responsible for different parts of the experience. For example, some models handle core reasoning and analysis, while others are optimized for lightweight tasks like dataset selection, naming or fast interactions.

Today, our production system primarily uses Google’s Gemini model family across these roles, allowing us to balance performance, speed and cost. This includes:

* A primary model for complex reasoning and user interactions
* Faster, lightweight models for simple or high-frequency tasks
* Dedicated models for code execution and data processing
* Embedding models for dataset retrieval and search

We continuously test and integrate models from other leading providers, including OpenAI and Anthropic, and may route specific tasks to different models as the system evolves.

This architecture allows us to improve reliability, efficiency and quality—while ensuring each part of the system is handled by the model best suited for it.

<mark style="background-color:blue;">**How it works:**</mark>&#x20;

1. **Processing your intent**: When you ask GNW Horizon a question, we use LangChain to process the natural language and determine your intent. This allows us to select the best AI models and analysis tools for the response.&#x20;
2. **Retrieving quality data**: Our data comes via APIs from Global Nature Watch (formerly Global Forest Watch)  and Land & Carbon Lab, initiatives powered by contributions from researchers and partners around the globe. This means peer-reviewed data from reliable sources.
3. **Composing a response**: The AI synthesizes the best response based the information at hand: user prompt, analyzed data and relevant context.&#x20;
4. **Returning a response**: Our agents are currently able to create spatial summary statistics, perform dataset searches and return natural-language summaries.