A conversation.
Not a dead end.
A useful chat agent needs more than a good answer. It needs business context, a clear next step and the judgement to bring a person into the conversation.
- FOCUS
- Conversational AI
- OUR ROLE
- Product & engineering
- BUILDING BLOCKS
- Knowledge, routing, handoff
- PROJECT TYPE
- Framework study
Answers aren’t the whole job.
A customer asking about a product and a customer reporting a damaged delivery may arrive through the same chat window. They need different responses—and different next steps.
Harbor is a web-chat agent framework for connecting those conversations to business knowledge and human support. The design challenge is knowing what to answer, what to retrieve and what not to automate.
Connect the knowledge.
Keep the conversation intact.
Harbor adds an agent layer around a conversation workspace. Retrieval provides business context; routing decides how to handle the request; escalation brings the team in when needed.
The underlying framework brings together Chatwoot for the conversation workspace, LangGraph for agent flow and a searchable business knowledge base.
Useful by design.
Bounded on purpose.
Context before confidence.
Retrieve the relevant business information. An articulate answer is not a substitute for a grounded one.
Handoff is a feature.
Make room for people when a request is sensitive or outside the agent’s scope.
Carry the story forward.
Keep the conversation context available to the team so the customer doesn’t have to start again.
Start with a real conversation.
The useful starting point isn’t “put AI on the website”. It’s understanding the questions customers ask, the information needed to answer them and the decisions your team should retain.