> For the complete documentation index, see [llms.txt](https://ungate.gitbook.io/ungate-infiniroute-avs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://ungate.gitbook.io/ungate-infiniroute-avs/architecture/abstract.md).

# Abstract

The high-level architecture of InfiniRoute is designed to efficiently route AI inference requests to the most suitable models, optimising for cost and accuracy. The system begins with the **Intelligence Consumer**, who initiates requests. These requests are managed by the **Chat Session Manager**, which coordinates user interaction and forwards the requests to the **Router**. The Router, a central component, determines the best model by consulting the **Models Library** for cost and endpoint information. Selected models, hosted externally, process the inference and return results.

For data management, session data is stored in the **Chat Session Store** and passed to the **AVS (EigenLayer)** for onchain storage. The **Node Operator** writes this metadata to the **IPFS**. **Attestors** validate the integrity of the metadata, which is then aggregated and written to the **L2 Blockchain**, with final storage on **Ethereum (L1 Blockchain)**. This architecture ensures seamless request handling, optimal model selection, and robust, transparent metadata management using blockchain technology.

<figure><img src="https://3183605978-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FSvxyExj0oUqpPY4D4VQC%2Fuploads%2FZ3hXonissv4HVg5o4bvV%2Fimage.png?alt=media&amp;token=f0d6c43f-491b-4909-9f24-59ad29a96b25" alt=""><figcaption></figcaption></figure>