Hi folks,
I’m pretty new to LangChain myself. Before working on this, I honestly didn’t know much about it.
As I started exploring it, I found myself wondering whether there are some cases where LangChain-style apps would benefit from voice interaction, not just text.
A few scenarios felt especially interesting to me:
- when the user is doing something else and can’t stay focused on a text UI
- when the interaction is naturally back-and-forth and more conversational
- when the app feels more like a live copilot than a chatbot
So I tried building a small integration pattern around that idea.
The basic split is:
- LangChain handles tool selection, tool execution, and response composition
- a real-time voice runtime handles RTC / RTM, speech input/output, and session lifecycle
- the two are connected through an OpenAI-compatible custom LLM endpoint
Here’s a screenshot of the current prototype:
What I found interesting is that this keeps the LangChain part pretty familiar. If someone already has LangChain-based tools or workflows, this feels more like adding a voice interface than replacing the whole app architecture.
The current implementation is intentionally simple:
- Python backend for the custom LLM bridge and agent lifecycle
- Next.js client for RTC + RTM interaction
- LangChain stays on the server side as the tool layer
I’d be curious what people here think:
- Does this split between “LangChain as the tool layer” and “voice runtime as the real-time layer” feel like a reasonable way to structure this kind of app?
- If this pattern is interesting, what would be the most useful next example: RAG, internal tools, docs copilot, or something else?