Cases / AI agents
Custom AI knowledge agent

Stabilo: an AI agent that makes company knowledge instantly available.

We built a custom AI knowledge agent for Stabilo that makes project documentation, manuals and additional knowledge searchable through an internal chat. Staff ask in plain language and get an instant answer, without digging through documents themselves.

Clientdestandbouwer.nl Typecustom AI agent ServicesRAG, AI chat, document processing Read time4 min
Stabilo AI agent for company knowledge
Knowledge sources project documentation, manuals and additional knowledge
Stack
RAG OpenAI Embeddings Vector search FastAPI
Since then in use by Stabilo staff, ready to expand to customer questions in the app
01

The challenge

Many companies already have a lot of knowledge in-house, but it sits across documents, manuals, project info and loose notes. The question is not whether the information exists, but how quickly someone finds the right answer.

For Stabilo we therefore did not build a generic chatbot next to the business, but an AI agent on top of their own knowledge. The agent helps staff without exposing internal processes and without letting sensitive information disappear into a loose AI tool.

mockup
stabilo · AI agent
smart_toy
AI knowledge agent
answers based on own knowledge
What do I need to know for this project?
1
I checked the linked project documentation and manuals. The key points:
checkuse the current project documentation;
checkcheck the additional knowledge for this project;
checkask a follow-up if you want to zoom in on a specific part.
Sources used
description
Project documentation
active source
2
menu_book
Manual
linked knowledge
lightbulb
Additional knowledge
added by Stabilo
Not a loose AI tool, but an agent that answers from the own knowledge base.
Screen 1 Internal AI chat. Staff ask in plain language. 1 The agent answers instantly, 2 based on linked sources.
02

What we built

We built a custom AI agent with a RAG setup: Stabilo fills the knowledge base with documents and additional information, and the agent makes that knowledge searchable via chat. The agent does not simply send entire documents to AI. First the right pieces of information are found; then the model only gets the relevant context to answer.

Mantella does not only deliver the agent, but stays responsible for maintenance, optimisation and further expansion. Because AI changes fast, we keep the agent sharp monthly and use new capabilities as soon as they add value.

mockup
knowledge base · documents
Fill knowledge base verifiedactive
upload_file Upload documents PDFs are processed, chunked and made available to the AI agent.
1
check_circle
Projectdocumentatie.pdf
processed · ready for AI
ready
check_circle
Handleiding 2026.pdf
processed · ready for AI
ready
lightbulb
Additional information
loose knowledge, corrections and additions
2
Screen 2 Fill the knowledge base. 1 Stabilo can upload documents themselves. 2 Additional knowledge does not need to sit in a PDF first.
mockup
RAG · from knowledge to answer
Knowledge in
descriptionProject documentation
menu_bookManual
lightbulbAdditional knowledge
hub AI knowledge agent finds the right pieces and composes the answer RAG · vector search
Answer out
smart_toyImmediately useful
A short, clear answer in plain language, based on the own knowledge base.
check_circlebased on own knowledge
Screen 3 From knowledge to answer. Documents and loose knowledge go in, the agent finds the right pieces, and a ready-to-use answer comes out.
Under the hood · how RAG works
01

PDFs are processed and split into text chunks.

02

Chunks are converted to embeddings.

03

Embeddings are stored in a secured knowledge database.

04

On a question, vector search finds the most relevant sources.

05

The AI generates an answer based on that context.

06

Chat context is carried over compactly for follow-ups.

API calls run server-side. The browser never sees keys.
03

What the agent does

Answers questions in plain language based on Stabilo's own knowledge
Searches linked documents, manuals and additional information
Lets Stabilo upload documents and add loose knowledge themselves
Supports multilingual answers and follow-ups with recent chat context
Is extensible: the same foundation can later answer customer questions in the Stabilo app
Is maintained, optimised and developed monthly by Mantella

The result

Less searching through manuals and documents
Faster answers to recurring questions
Knowledge centrally managed, in one place
Ready to grow into customer questions in the app
Mantella insight

An AI agent only becomes valuable when it does not sit loose next to your business, but works on top of your own knowledge. For Stabilo we therefore did not build a generic chatbot, but a knowledge layer that turns documents, manuals and additional information into ready-to-use answers.

M
Mantella · custom AI agents
Get in touch