A chatbot over your documents (RAG)
Searches by meaning and by exact word across your documents and cites the source. Measured before it reaches production.
AI in the cloud · Azure, AWS and Google Cloud
Artificial intelligence inside the cloud you already pay for
If your company already runs on Microsoft Azure, Amazon Web Services or Google Cloud, the models, the indexes and the data can live in your own account. We build document chatbots, agents and applications with your credentials and the usage billed in your name.
- 3 years
- running Azure OpenAI and Azure AI Search in production
- 2M+
- users served by that system across ten countries
Your cloud account
- Microsoft Azure
- Amazon Web Services
- Google Cloud
Your permissions
Your documents Search AI model Agents
Nothing leaves your cloud
In short
- With Azure OpenAI, Amazon Bedrock or Vertex AI, what is sent to the model is not used to train it, and the search index lives in the same account as the rest of the data.
- The architecture is the same on all three clouds —search your documents, answer with the source, measure before launch—; only the service names change.
- Quarl has run Azure OpenAI and Azure AI Search in production for three years, in a RAG system for two million users across ten countries. On AWS and Google Cloud we start with a scoped pilot and measure it the same way.
- Resources, keys and usage stay in the company’s name, billed by its provider, with no Quarl account in the middle.
-
Microsoft Azure
- AI models
- Azure OpenAI, in Microsoft Foundry
- Search over your documents
- Azure AI Search
- Permissions
- Entra ID
- Who bills
- Microsoft, in your name
-
Amazon Web Services
- AI models
- Amazon Bedrock: Claude, Llama and Mistral
- Search over your documents
- Bedrock knowledge bases or pgvector
- Permissions
- IAM roles
- Who bills
- AWS, in your name
-
Google Cloud
- AI models
- Vertex AI: Gemini and Claude
- Search over your documents
- Vertex AI Search
- Permissions
- Cloud IAM
- Who bills
- Google, in your name
What security asks
Where does the data go?
- The model, the index and the logs in your account
- The permissions you already manage
- Usage billed by your provider, in your name
What we build
The pieces, inside your account
The right model for each task
A cheap one classifies and summarizes, a capable one writes the answer. Cost per query in plain sight.
Agents that query your systems
With permissions scoped to what each tool needs and human approval on steps with consequences.
Input and output filters
They block topics and personal data, on top of the defenses against malicious instructions.
Formats
Scopes and timelines
-
4–8 weeks
Document chatbot in your cloud
Ingestion, index, a model from your cloud citing the source and an evaluation set, deployed in your account.
-
2–4 weeks
Move a pilot into your cloud
A prototype built on a provider’s public API, moved into your cloud with quality measured before and after.
-
1 week
Diagnosis of what you already have
Search and answers measured separately, cost per query and the failures ranked by impact, each with its fix.
When not to hire us
If you are on none of these clouds, we will not move you for this.
The same architecture runs on the provider’s direct API. If what you want is for the team to draft emails and summarize meetings, Microsoft 365 Copilot or Gemini in Google Workspace already does it. And if the work is the infrastructure itself —networking, landing zone, server migration— we are not the team: we build what runs on top.
Related services
Chatbots for web and WhatsApp
With the evaluation set run before it ever reaches production.
View serviceAI agents
LangGraph orchestration, durable state and human approval on steps with consequences.
View serviceCustom software
The system around the model, not just the model.
View serviceAI consulting
Architecture, evaluation criteria and cost per query before writing code.
View serviceQuestions about AI in the cloud
If yours is not here, that is what the fifteen-minute call is for.
Is our data used to train models?
No. Azure OpenAI, Amazon Bedrock and Vertex AI do not use what is sent to them to train their models. The search index, the documents and the logs stay in resources in your account, under the permissions you already manage. We also filter personal data before it reaches the model when the use case does not need it.
Which cloud should we use?
The one you already use. It is not a better model: it is the contract, the identity and the bill the company already has. On Microsoft, Azure; if the operation lives on AWS, Bedrock; on Google Cloud, Vertex AI. Moving clouds for an assistant is not justified.
What experience do you have on each cloud?
Most of it is on Azure: three years running Azure OpenAI and Azure AI Search in production, in a RAG system for two million users across ten countries. On AWS and Google Cloud the architecture and the measurement are the same; there we start with a scoped pilot and measure it the same way before growing.
Who pays for cloud usage?
You do, directly to your provider, within your account or enterprise agreement. Quarl does not resell usage or put an account of its own in the middle. What we do deliver is the measured cost per query, so the bill is not a surprise when usage grows.
Can you work on what we already have?
Yes, and it is the most common case: an index or a deployment that already exists and answers badly. We start by measuring it with real questions and their correct answer, as in a rescue, and the fix is made on your resources, without rebuilding what works.
Fifteen minutes. A concrete answer.
It gets fixed, it gets built, or it is not worth it. And if the diagnosis does not reach three actionable findings, it is free.