Architecture

Overview

Index enables storing vector embeddings and indexed data on a decentralized graph, supported by ComposeDB, allowing autonomous agents to generate and search these embeddings via the Fluence Network. Users interact through natural language queries for personalized responses or subscribe to contexts, triggering agents on updates, all within a decentralized, privacy-focused environment.

Decentralized Semantic Index

Index enables discovery through a composable vector database, which stores vector embeddings on the same graph with the indexed data. Ceramic Network’s decentralized knowledge graph solution, ComposeDB, supports this approach, allowing modularity for a highly relational composition of use cases.

In Index, autonomous agents generate vector embeddings using different AI models from multiple perspectives, such as entities, facts, intents, or summaries. Each embedding is fed to a search service on the Fluence Network to enable fast querying, tiered in different privacy settings within user/agent-owned environments.

Index offers two main methods to interact with indexes:

1. Natural Language Queries Across Multiple Indexes: For example, within a chat setting, responses can be generated using both users' intent index and a community index. Even other indexes can be indexed. This enables users to have a connected discovery experience, ensuring that responses are both personalized and trusted.

2. Contextual Pub/Sub: Index enables a “contextual pub/sub,” allowing users to subscribe to contexts using natural language queries, such as simply saying, "run this agent if something new happens about quantum computing."

Event-Driven Agent Runtime

In Index, each query is answered using perspectives from dozens of independent agents. To make this possible, agents are continuous, living entities that function as real-time event listeners from the semantic index. This creates a contextual environment for agents, allowing them to discover and coordinate with other agents using natural language. Alongside built-in agents, the decentralized runtime enables users to dynamically schedule agents that can be generated via natural language. All agents operate within per-user environments on a decentralized compute network, ensuring complete privacy.