ServicesRAG development
RAG development services for answers from your company documents
Your staff get answers from manuals, policies, contracts and past tickets in seconds, with the source shown, instead of searching folders or interrupting the colleague who knows. RAG, short for retrieval-augmented generation, finds the relevant passages in your documents and hands them to a language model before it answers. We build these systems inside your environment with your team, or repair ones you already run. They suit teams in service, support, sales and administration who answer questions from documents every day. Our engineers sit in Germany and Poland and test in English, German and Polish.
How a question gets answered
- 1
Your sources are ingested with their owner, version and access rules.
- 2
Search combines exact identifiers with retrieval by meaning.
- 3
Only passages the person asking may read reach the model.
- 4
The answer cites its sources. Without support in them, the system says so and, if a handover is set up, passes the question to a person.
What changes for your business
Knowledge stops depending on who you happen to be able to ask. A support engineer, a new colleague in the back office or a sales rep asks in their own words and gets an answer from your approved documents in seconds, with a link to the passage. The experts who used to answer the same questions every week get that time back for the cases that need them.
- Answers in seconds instead of searching folders, wikis and old tickets
- Fewer interruptions for the experts who know where everything is written down
- New staff productive sooner, because they can look things up on their own
- Customers answered faster, because service staff find the answer during the call
- Answers you can check, because each one cites its source
More detail
Companies are moving this way. In Bitkom's 2026 survey, 22% of companies that use AI use it for internal knowledge management, up from 11% in 2025. What this looks like for a whole department, from the first questions to the works council, is on our knowledge management use case page.
An example of the work
Suppose a support engineer asks which firmware versions of a device need a manual reset after an update. The answer sits in two approved runbooks and one release note. The system has to find the release note by its exact version number, pick the runbook section that applies and answer with a link to each document. If the release note was withdrawn last week, it must not be used.
Sources, owners and permissions
We start by reviewing your sources and who may access them. Every passage keeps a reference to its original document, owner and version, so an answer can be traced back. After that review we connect to the systems where the documents already live.
More detail
Permissions are checked before any text reaches the model. Someone who cannot open a document in its original system should not get an answer built from it.
- Access matched to the permissions of the original system
- Edits and deletions synchronised, so removed content leaves the index
Would you like to talk this through for your own case? The first call is free and takes 30 minutes.
Book a 30-minute callFinding the right passage
Documents are split into chunks for search, and a chunk cut out of its document can lose the context that gave it meaning, so we keep that context with it. Search by meaning copes well with everyday phrasing but can miss exact strings such as error codes or part numbers. Where those matter, we add lexical search.
More detail
Each answer cites the passages it used. When nothing relevant turns up or the sources disagree, the system says so and can hand the question to a person.
Tests, operation and handover
Before launch we build a test set with your team. Retrieval and answers are scored separately, because a better prompt cannot fix a passage that never reached the model. We also track latency and cost per question.
More detail
When the system goes live, your team receives the tests, the sync jobs and a runbook for adding sources, changing permissions and investigating a wrong answer.
- Questions the documents answer, with the expected sources
- Questions they cannot answer, where the system should decline
- Questions whose answer changed in a newer document
- Questions from users without access to the source
When a custom RAG system is worth building
Some cases need less. Short documents that rarely change can go to the model directly. If your existing search already finds the right document, improving it may be enough, and a managed knowledge base product covers common sources with simple permissions. When one of these fits, we tell you in the first call.
More detail
Fine-tuning changes how a model behaves, for example its output format or how it classifies a request. Facts that change every week belong in retrieval, since otherwise each change would need a new training run.
Questions
Does a knowledge assistant pay off for us?
It pays off where many people look up the same kinds of information every day, or where a few experts spend much of their week answering questions. We take a sample of real questions, note how long each takes to answer today, and measure again with the system on the same questions. Whether you roll it out further depends on that comparison.
What drives the cost of a RAG system?
The number and condition of the sources, how complex the permissions are, the languages and document formats, whether it runs on your own servers or in your own cloud account, and how many people use it. Model usage is a running cost per question, which we measure from the start. Each step has a fixed price, agreed in writing after the first call.
Do our documents stay in the company?
They can. The index, the search and the language model can all run on your own servers or in your own cloud account, with an open-weight model, one whose weights you can run yourself, if no external provider may see the text. What is stored, where it is processed and who can read the logs is documented for your data protection officer, who decides with your lawyers.
What happens in the first call?
Thirty minutes with one of our engineers, free of charge, about the questions your team needs answered and where the documents live. We need no access to your systems. Afterwards we propose a scope and a price.
Can you repair a RAG system we already run?
Yes. We run it against a test set built from real questions and score retrieval and answers separately. That shows where the problem sits and which fix to start with.
Will the system stop giving wrong answers?
No system is right every time. Citations, tests with unanswerable questions and the option to decline reduce wrong answers. The test set shows how often it gets your questions right.
How many documents do we need?
There is no fixed minimum. What counts is whether the documents contain the answers your users look for and whether each source has an owner. We check both in the source review.
Related services
Data integration for AI
Answers from file shares, SharePoint, ERP and tickets in seconds, for assistants and agents, with permissions carried through and tested.
AI agents and process automation
Agents that take on the routine steps of complex cases and pass decisions to people, with a cost per handled case.
AI chatbots
Customers and staff answered faster from your data, with approved actions and a handover to people when it matters.
System access for agents (MCP)
AI agents that look up and update data in your systems, with narrow permissions, human approval for changes and tests.
Custom model training
A small open-weight model trained for one task and compared with your current model on cost and accuracy.
On-premise AI
AI on offers, contracts and customer files, with the data staying in your company: own servers or own cloud account, tested before go-live.
AI system testing
Before go-live, acceptance tests on your own cases tell you how often your chatbot, assistant or agent is right, in terms your QA team already uses.
From the blog
RAG document permissions
How access rules travel from source systems into a retrieval index, where they get lost, and which negative tests show that a user cannot retrieve what they may not open.
Retrieval or generation?
A diagnostic procedure for finding out whether a wrong RAG answer started in ingestion, search, reranking or the model, using labelled evidence and per-question traces.
Stale documents in RAG
How to keep a retrieval assistant current when documents are edited, renamed, restricted or deleted, with a sync design and canary tests that prove old content is gone.
Talk to an engineer
Bring a few questions your team spends too long answering from its documentation. The call takes 30 minutes and needs no access to your systems or your data.
Booking runs on Google Calendar.
When you book a call, you speak to Mickey Graf
He founded Cloudsail and has put machine learning into production since 2017, including AI engineering at trivago and SAP. The call can be in English, German or Polish.
Prefer to write first? Leave your email and a few lines. Mickey reads every message himself and replies by email with times for a call.
Look for the draft in your email program
Send it from there, and Mickey replies in person with times for a 30-minute call. If no draft has opened, write to us at miki@cloudsail.com.