Building a Decentralized Multi-Agent Travel Assistant with Fetch.ai, LangGraph, MCP Tools, Chat Protocol and ASI1 LLM

Syed Zakiyuddin — Fri, 06 Jun 2025 15:27:28 +0000

In an effort to demonstrate the potential of agent-based systems in real-world use cases, we developed TravelBud, an intelligent, decentralized travel assistant built entirely within the Fetch.ai ecosystem. TravelBud leverages the uAgent chat protocol, the LangGraph framework, MCP Tools and the ASI1 large language model (LLM) to orchestrate a seamless, conversational travel experience powered by autonomous agents.

🧭 Motivation

Traditional travel planning is often fragmented and centralized, involving multiple platforms and intermediaries. Our goal was to build an all-in-one, privacy-respecting, modular, and intelligent AI system where autonomous agents interact, reason, and respond intelligently without centralized servers.

⚙️ System Architecture

1. Multi-Agent Workflow with LangGraph

We used LangGraph to design a graph-based workflow of agents. Each node in the graph represents a distinct skill:

Flight Search
Hotel Search
Tourist Spot Finder
Budget Estimator
Travel Tips Generator

Basically a complete travel itinerary.

LangGraph’s flexibility enabled dynamic path execution based on user intent, enabling natural conversations with powerful task delegation.

2. uAgent Adapters for Fetch.ai Ecosystem

To expose our LangGraph agents to the uAgent network, we built custom adapters that wrapped LangGraph tools into uAgent-compatible services. These adapters:

Handled TextMessage payloads
Managed message parsing and tool invocation
Supported response formatting for final delivery

Each agent was deployed on the Agentverse, making it publicly discoverable and interoperable with other uAgents.

3. ASI1 LLM as the Central Brain

We integrated the ASI1 LLM, Fetch.ai’s language model, to act as the primary reasoning engine. It handles:

Interpreting natural language from users
Mapping intent to LangGraph tool invocations
Generating multi-turn, contextual responses

This allows TravelBud to act less like a rigid bot and more like a fluid, goal-directed assistant.

💬 Custom Chat Protocol

A key innovation in our project was the design and implementation of a custom chat protocol to enable real-time, asynchronous messaging between agents and users within the Fetch.ai ecosystem.

Here’s how it works:

Messages are exchanged using TextMessage events, the standard communication primitive in uAgent.
Incoming messages are routed through a central handler that:
- Detects user intent using ASI1
- Triggers relevant LangGraph tool calls
- Captures and formats the result for return
Each tool’s output is structured and sent back through the uAgent message relay
All interactions maintain state via LangGraph memory, allowing multi-turn context-aware dialogue

This protocol allows for low-latency, modular, and fully decentralized chat experiences—without needing centralized servers or third-party integrations.

✅ Highlights

🧠 Modular Reasoning with ASI1 and LangGraph
🌐 Fully Decentralized via uAgent chat protocol
🔁 Multi-Turn Memory & Intent Tracking
📡 Real-Time Chat Protocol for agent communication
MCP Tool Integration

🧪 Learnings

Agent communication across ecosystems (LangGraph → uAgent) requires thoughtful serialization and routing
Building chat protocols demands managing both state and logic across tools and transports
ASI1’s LLM reasoning capabilities scale well when modularized into LangGraph tools

🚀 What’s Next?

We’re exploring:

Inter-agent negotiation for group travel
Integrating decentralized identity and payments
Extending the chat protocol to support media and location-rich messages

🗺️ Explore the Project

📄 TravelBud: Travel AI using Fetch.ai

TravelBud shows how powerful, composable, decentralized AI systems can be built today using the Fetch.ai ecosystem. Whether you're building personal assistants, supply chain bots, or healthcare agents this stack is production-ready.

DEV Community: Syed Zakiyuddin