AI Chatbot Faster Than Custom Builds, More Flexible Than SaaS
1–2
Months to production
From kickoff to live, depending on integration and branding complexity.
4–5
People in a typical build
An AI engineer, a DevOps engineer, a QA engineer and a PM, with design support for custom branded experiences.
MIT
Open-source licence
Available on GitHub, with no licensing costs and no vendor lock-in.
The client
Netguru’s R&D team built Chatguru in-house: an MIT-licensed, open-source platform for AI chatbots. Being white-label, it hands a team a working foundation rather than an empty repository.
The platform covers the backend, the conversational logic and the AI infrastructure; the interface components come from Silk, the free design system Netguru publishes for commerce products. The two behave like modular blocks a team shapes to its own brand and workflows.
The challengeWhat the teams actually needed
The chatbot market offered two options, and neither fitted teams trying to build conversational experiences grounded in real business data.
Challenge 01SaaS tools hit a ceiling
Quick to launch, but customising flows or connecting catalogues, inventory and internal tools meant workarounds or dead ends, with vendor lock-in as the default outcome.
Challenge 02Custom builds cost months
Full control came at the price of months of engineering on RAG, architecture, observability and testing before anyone could get to the business value.
Challenge 03Reliability as a business risk
Healthcare use cases needed answers that stayed compliant and traceable to verified sources; commerce needed answers matching real availability and live data.
What we did
A production-ready foundation, not a prototype
Every architectural decision targets the reliability, flexibility and implementation problems teams were hitting with the existing options.
Answers grounded in the client’s own data
Instead of leaning on generic model knowledge, Chatguru retrieves from the client’s product catalogues, policy documents or knowledge bases before it generates a response.
Self-hosted, so data stays put
Client data never leaves the client’s own infrastructure, which makes the platform usable in regulated industries, and the MIT licence removes vendor dependency.
Paired with Silk
Chatguru owns the conversational logic and AI infrastructure; Silk supplies reusable frontend components. A branded experience therefore does not mean building both ends from scratch.
Monitoring and evaluation by default
The default setup ships with Langfuse tracing, pytest, Promptfoo for evaluating the LLM, and RAGAS for measuring RAG quality.
Tech stackThe tools that made it possible
01
FastAPI & Uvicorn
The backend, chosen for maintainability in real production environments rather than demos.
02
LangChain & Azure OpenAI
The default LLM layer.
03
sqlite-vec
Vector search behind the retrieval layer.
04
React 19, Vite & Docker
The frontend stack, with the full stack launched in one command through Docker Compose.
Netguru quote
We built Chatguru because we kept seeing the same problem: teams either accepted the limitations of SaaS tools or spent months building AI infrastructure from scratch before getting to the actual product. Chatguru is a production ready foundation that’s genuinely adaptable to whatever the business needs it to do
More like thisOther places we scaled product teams for growth.
What’s next