StayFlow
An agentic AI travel assistant that helps travellers discover, plan, and book trips through conversation.
- 01
The situation
Travel planning often means jumping between different websites and tools.
Travellers may need to search for hotels, understand destinations, compare packages, find things to do, check weather, locate places on a map, and eventually make a booking.
Traditional travel websites handle these tasks separately, while generic AI assistants may answer questions without being connected to the actual travel services or information needed to complete the journey.
StayFlow was built to bring these interactions together into one conversational travel experience.
- 02
What we found
Travellers don't always know exactly what they need to search for.
A simple question such as “Where should I stay?” can lead to several follow-up needs:
Which hotels are available?
What's included?
What can I do nearby?
What's the weather like?
Where is the place?
Can I see it on a map?
Can I book it?
The challenge wasn't simply generating better AI answers.
StayFlow needed to understand the traveller's intent, retrieve relevant information, use the appropriate tools, and take action when required.
- 03
What changed
Hotels
Travellers can ask about hotels, properties, amenities, locations, packages and other available information through natural conversation.
Destinations
The agent can help travellers explore destinations, attractions, activities and places based on the available travel information.
Packages
Travellers can ask what's included in a package, which destinations it covers, what experiences are part of it, and other details without manually going through separate pages.
Experiences
The agent can help discover things to do based on the destination, package and traveller's conversation.
Weather
Weather information can be retrieved through connected tools when a traveller needs current conditions or wants to understand what to expect at a destination.
Locations & Maps
When a traveller wants to find a hotel, attraction or other place, StayFlow can provide location information and relevant map links.
Booking
The experience can move beyond recommendations into action. When booking functionality is connected, StayFlow can help the traveller select an option and complete the booking directly through the conversation.
Tool use
StayFlow doesn't treat every request as a text-generation problem. The agent determines when it needs additional information or an external tool — retrieving travel data, checking information, using location or weather services, or initiating a booking flow.
Grounded responses
Travel-specific answers are generated using relevant information from the connected travel knowledge base and services, rather than relying solely on the model's general knowledge.
Query understanding
The agent interprets natural, incomplete or conversational requests and turns them into useful retrieval or tool actions.
Multi-turn conversations
The conversation maintains context, allowing travellers to refine their plans naturally:
“Find hotels in Srinagar.”
“Which ones are near Dal Lake?”
“What will the weather be like?”
“Show me the location.”
“Book this one.”
Streaming responses
Responses are streamed as they are generated, creating a more natural, interactive experience.
- 04
Result
StayFlow became an agentic travel experience rather than a conventional AI chatbot.
Travellers can use one conversation to discover hotels and destinations, explore packages and experiences, check weather, find locations, open maps, retrieve relevant information, and complete bookings through connected tools and services.
The result is a travel assistant that doesn’t just answer questions — it can understand what the traveller is trying to accomplish and help carry out the next step.
Behind the build
- Agentic orchestration
The system coordinates conversation, knowledge retrieval and external tools based on the traveller's request, allowing different types of actions to happen within the same interaction.
- Travel knowledge & retrieval
Travel information is structured into a searchable knowledge layer. Relevant information is retrieved based on the traveller's question and supplied to the model as context.
- Tool integration
External capabilities are connected to the agent for tasks such as weather, location, maps and booking. The agent can determine when these capabilities are relevant instead of treating every request as a simple knowledge query.
- Grounded generation
Retrieved travel information and tool results are incorporated into the response so that travel-specific answers are based on the information available to the system.
- Booking workflow
Booking functionality connects the conversational experience with the underlying reservation flow, allowing the traveller to move from discovery to selection and booking without leaving the experience.
- Conversation context
Previous messages are maintained across the interaction, allowing the agent to understand follow-up requests and progressively build context around the traveller’s plans.
- Streaming & response handling
Responses are streamed to the interface while the application manages conversation state, tool calls and the resulting information.
- Travel-focused architecture
The system was designed around the actual travel journey — discover → understand → decide → act — rather than simply placing a chatbot on top of existing travel content.
Next in the series
ForgePlaceHave something worth building?
Whether you're improving something that already exists or creating something new, let's see where it can go.