What we do · Private AI & RAG
We set up and extend private AI assistants on your own documents: self-hosted, connected to your sources, with every answer pointing to its source.
In short
Private AI & RAG integration is a service by Mandala Labs GmbH, Berlin, for organizations that cannot hand their documents to a third party. We set up AI assistants that answer from your own documents using retrieval-augmented generation (RAG), on your infrastructure, based on our self-hosted AIDocumentIndexer and extended with custom pipelines, agents and your choice of AI models.
What we do
Four building blocks. Combine them as your requirements call for.
We install it on your servers or in your cloud account and connect it to where your documents live. Staff ask in plain language and get answers with a citation to the exact passage.
40+ connectors · 30+ formatsWhen a standard set-up does not fit your data or questions, we build the retrieval around them, so the model works from your own material.
Built to your sourcesAgents that run on a schedule or when new documents arrive, and workflows with approval steps, so a person signs off before anything is acted on.
Approval steps · replayable runsLocal models on your own hardware, your own cloud, or an API provider you pick. With fully local models, no document content leaves your network.
Local · own cloud · APITypical results
With citations to the source, running on your infrastructure.
No more guessing file names across drives, inboxes and PDFs.
Recurring tasks triggered by new documents, with an approval step before results are used.
Generic examples, not client results. Concrete outcomes depend on your documents and questions. Our reference is our own product, AIDocumentIndexer.
How we work
We map which documents, systems and questions matter, who may see what, and which hosting and model options your rules allow.
A first set-up on a sample of your documents, so your team can try real questions and judge the cited answers.
We connect your sources, set up access rules and add custom pipelines, agents or workflows where needed.
Roll-out to your team, then updates and model changes as you need them. We run it for you or hand it over to your team – as you prefer.
The basis
A private AI assistant for your organization’s documents: self-hosted, plain-language, every answer cited. Where it fits, a project starts from it instead of a blank page.
FAQ
Something missing? Write to contact@mandala-labs.com.
Retrieval-augmented generation (RAG) lets a language model answer from your own documents: relevant passages are retrieved first and the answer is based on them. “Private” means the system runs on infrastructure you control, so your documents are not handed to a third party.
Not with local AI models: AIDocumentIndexer runs on your own infrastructure and, with fully local models, no document content leaves your network. If you choose a cloud model provider, queries go to that provider.
Models on your own hardware, or a provider you choose, such as OpenAI, Anthropic, Google Vertex, AWS Bedrock or OpenRouter. Hosting is on your own servers or in your own cloud account. In AIDocumentIndexer, switching the model is a settings change.
A self-hosted setup keeps documents under your control, and AIDocumentIndexer provides access tiers, per-collection allow and deny rules, invitation-only registration and an audit log with CSV export. These controls support your own GDPR compliance work.
A short description of your sources and questions is enough to start. We reply with how we would approach it.
contact@mandala-labs.com
