Technology Architecture
ELNA technical architecture
The technical architecture of ELNA is designed to leverage the Internet Computer (ICP) canister architecture. ELNA supports both on-chain and off-chain tools and LLMs for practicality and data is stored in the on-chain vector database in the user canister.
On-chain we have the following functions
Agent Framework
Retrieval Augmented Generation orchestration with Vector DB and Similarity Search
User Interface
User Management
On the client side, we have data processing and wallet login
In the form of external services (off-chain), we have
offchian vectorisation,
support for inference engines like chat gpt and LLAMA,
integration tools to X, telegram, websites etc
Our longtime goal is to minimise the use of external/off-chain components and move completely on-chain on the ICP blockchain. Due to current limitations of instruction limits, memory limits and availability of GPU we need to support some off-chain AI components for practical applications. Our team will be committed to developing and evangelising multiple AI development tools like on-chain inferencing, fine-tuning support, benchmarking etc
On-Chian Canister Architecture
We use multi-canister architecture with the following components in ELNA platform canisters and data which are stored as a vector in the user canister.
On ELNA Platform
We have the canisters for user management, orchestration of agents, marketplace, developer management and we have a vector canister which is created on the user side.
NB: currently multiple canisters are combined, in the coming version the splitting will be done and those will be added to the SNS
User Management Canister: Central to handling the diverse user base, this canister manages the roles, permissions, and profiles of users. It ensures secure authentication and authorization processes for different user types within ELNA, from developers to end-users.
ELNA Dapp Canister: This canister encapsulates the core functionalities of the ELNA decentralized application. It includes the user interface, facilitating interactions with the platform, and managing connections to other canisters for a cohesive user experience.
Agent Management Canister: Focused on the lifecycle of LLM agents, this canister lists agents' details such as biographies, prompts, descriptions, and tools used in their creation. It also houses marketplace information, including pricing, ratings, and usage statistics, facilitating the discovery and utilization of agents.
Developer Management Canister: This canister supports developer engagement by managing accounts and providing access to essential tools and resources. It streamlines the listing of tools such as inference engines, LLMs, data sources, and monetization avenues, fostering an ecosystem for AI development.
Prompt Engine Orchestration Canister: Key to enhancing LLM interaction, this canister orchestrates the prompt engine, linking it with the vector database for efficient data retrieval and processing. It plays a crucial role in the generation and handling of prompts for LLMs.
Log Canister: Dedicated to operational transparency and management, the log canister records all usage data. This information is vital for administrative purposes, including monitoring, analytics, and troubleshooting.
Agent Vector DB Canister: At the core of ELNA's data handling capabilities, this canister stores user and agent data as vector embeddings. This innovative approach facilitates advanced data retrieval and similarity searches, essential for the platform's AI-driven functionalities.
Type of Operators
User
In the ELNA ecosystem, a User is an individual or entity that interacts with the platform to leverage the capabilities of large language model (LLM) agents. They are the primary beneficiaries of Elna's offerings, utilizing the on-chain prompt engine, vector database, and integration with on-chain and off-chain LLM inference engines for various applications. Their interaction with the platform, governed by the use of Elixir tokens, drives the demand for ELNA's services, thus contributing to the ecosystem's value. As ELNA transitions into a Decentralized Autonomous Organization (DAO), Users also contribute to the governance model, proposing and implementing improvements.
Creator
Creators on ELNA are individuals or groups who design, train, and deploy LLM agents on the platform. They can range from individuals to businesses like Sonic Dex utilizing ELNA’s decentralized infrastructure to contribute innovative and diverse agents to the ecosystem. Creators play a pivotal role in expanding the platform's capabilities and offerings. With plans to launch a marketplace for user-generated LLM agents, creators are poised to have monetization opportunities, enabling them to earn revenue from their contributions. By burning ELNA to obtain Elixir tokens, creators can deploy their LLM agents on the platform, fostering a vibrant and innovative ecosystem of decentralized intelligence.
Developer
Developers in the ELNA ecosystem are the technical backbone, engaged in the development, maintenance, and enhancement of the platform itself. This includes blockchain developers, machine learning experts, and other technical specialists who work on the underlying infrastructure that enables on-chain operations, seamless integration of LLM agents, and the overall functionality of the ELNA platform. Their expertise ensures the platform remains at the forefront of privacy, transparency, and efficiency in decentralized artificial intelligence.
Last updated
