From static websites to conversational knowledge
We are no longer only searching for information — we have started talking to it. The shift is technical, behavioural and strategic, and it is moving faster than many organisations are ready for.
- RAG
- LLM
- Egen drift
Expectations have changed. People do not want to learn how your website works. They expect it to learn how to help them.
That is a behavioural shift, not a design trend. Conversational interfaces are not an improved user experience — they are a new form of digital interaction, and users will expect it everywhere.
What we tested
llms-txt is an open tool for building conversational agents on your own content — text files, web content, documentation — without external APIs or cloud services. It runs locally with Ollama and gets going quickly. Think of it as a minimum viable RAG system, ready to talk.
How it compares
It is not alone, of course. The RAG ecosystem has several good options, and the choice depends on stack, budget and use case:
- LangChain — complete and modular, for production pipelines
- Haystack — enterprise-ready, with a strong community
- llama-index — built to structure and index large document collections
- PrivateGPT — privacy-first, with local inference
- Claude, ChatGPT and Perplexity — easy to use, but tied to the cloud
- Flowise, Qdrant, Weaviate and FAISS — components for custom architectures
The value of llms-txt is the low barrier: you can test a hypothesis against real content quickly, without overengineering anything.
Why this matters
Language models and conversational interfaces can make knowledge more accessible, particularly where the content is complex or overwhelming: regulations, benefits, education, migration, finance. The information is rarely missing. It is that it is impossible to get through.
For anyone wanting to use language models in a product or internal workflow without handing the data to the cloud, local operation is worth a closer look. It is not a compromise — it is often the only defensible option.