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
TextMessagepayloads - 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
TextMessageevents, 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.
Top comments (0)