ReadMe.md

Vector Agent Routing Module

This module is part of the Intelligent Agent Routing System and focuses on resolving agent routing specifications using vector embeddings. It provides auto-configuration for Spring Boot applications to easily integrate vector-based agent routing.

Overview

The Vector Agent Routing Module uses vector embeddings to represent queries and agent capabilities, comparing them using cosine similarity. This approach is efficient for large-scale data and is independent of external APIs.

Features

• Vector Search Client: Searches for similar vectors in a vector store.
• Vector Seed Client: Seeds the vector store with agent routing specifications.
• Agent Routing Specs Provider: Reads agent routing specifications from a JSON file.
• Vector Agent Routing Specs Resolver: Resolves agent routing specifications using the vector search client.

Quickstart Guide

Step 1: Add Dependencies

Ensure you have the necessary dependencies in your build.gradle or pom.xml file.

dependencies {
    implementation("org.eclipse.lmos:lmos-router-vector-spring-boot-starter:1.0.0")
}

Step 2: Configure Properties

Set the required properties in your application.yml or application.properties file.

route:
  agent:
    vector:
      specFilePath: "path/to/your/agent-specs.json"
  llm:
    vector:
      search:
        threshold: 0.5
        topK: 1

Step 3: Initialize the Vector Agent Routing Specs Resolver

The module provides auto-configuration, so you only need to inject the VectorAgentRoutingSpecsResolver where needed.

import org.eclipse.lmos.router.vector.VectorAgentRoutingSpecsResolver
import org.springframework.beans.factory.annotation.Autowired
import org.springframework.stereotype.Service

@Service
class AgentRoutingService {

    @Autowired
    private lateinit var vectorAgentRoutingSpecsResolver: VectorAgentRoutingSpecsResolver

    fun resolveAgent(query: String): String? {
        val context = Context(listOf(AssistantMessage("Hello")))
        val input = UserMessage(query)
        val result = vectorAgentRoutingSpecsResolver.resolve(context, input)
        return result.agentName
    }
}

Configuration

Properties

• route.agent.vector.specFilePath: Path to the JSON file containing the agent routing specifications.
• route.llm.vector.search.threshold: Similarity threshold for vector search (default: 0.5).
• route.llm.vector.search.topK: Number of similar vectors to return (default: 1).

Beans

The module provides the following beans:

• VectorSearchClient: Uses the vector store to search for similar vectors.
• VectorSeedClient: Uses the vector store to seed vectors.
• AgentRoutingSpecsProvider: Reads agent routing specifications from a JSON file.
• VectorAgentRoutingSpecsResolver: Resolves agent routing specifications using the vector search client.

Usage

Seeding the Vector Store

To seed the vector store with agent routing specifications, use the VectorSeedClient.

import org.eclipse.lmos.router.vector.VectorSeedClient
import org.eclipse.lmos.router.vector.VectorSeedRequest
import org.springframework.beans.factory.annotation.Autowired
import org.springframework.stereotype.Service

@Service
class VectorSeedService {

    @Autowired
    private lateinit var vectorSeedClient: VectorSeedClient

    fun seedVectors() {
        val seedRequests = listOf(
            VectorSeedRequest("Agent 1", "This is the description for agent 1."),
            VectorSeedRequest("Agent 2", "This is the description for agent 2.")
        )
        vectorSeedClient.seed(seedRequests)
    }
}

Resolving Agents

To resolve agents based on user queries, use the VectorAgentRoutingSpecsResolver.

import org.eclipse.lmos.router.vector.VectorAgentRoutingSpecsResolver
import org.springframework.beans.factory.annotation.Autowired
import org.springframework.stereotype.Service

@Service
class AgentRoutingService {

    @Autowired
    private lateinit var vectorAgentRoutingSpecsResolver: VectorAgentRoutingSpecsResolver

    fun resolveAgent(query: String): String? {
        val context = Context(listOf(AssistantMessage("Hello")))
        val input = UserMessage(query)
        val result = vectorAgentRoutingSpecsResolver.resolve(context, input)
        return result.agentName
    }
}