Mandu'a
Mandu'a is an AI application that captures a company's emails, WhatsApp messages, documents and system data, organizes them by process and lets you query everything in natural language. We run the LLM infrastructure in the cloud (Azure) — models served via vLLM/TGI, multilingual embeddings and a vector database for semantic search — with per-client data isolation.
Built for document- & deal-heavy teams
Document-heavy companies drown in scattered correspondence. Nothing is connected, nothing is findable — and when people leave, the context leaves with them.
Emails buried in personal inboxes nobody else can see.
Deals negotiated across WhatsApp groups, lost to scroll.
Contracts & invoices scattered across desktops and drives.
Scanned PDFs that no search can read.
Four things happen automatically — no dragging files around, no manual tagging.
Reads a dedicated mailbox, linked WhatsApp groups, document folders, meeting transcripts and pushed database records.
Files everything by business process into structured, auto-numbered folders with navigable timelines.
Extracts text from PDFs & Office, runs OCR on scans, and pulls the real issue date from inside each document.
Hybrid search with rich filters, plus AI chatbots that answer in natural language and always cite their sources.
Each company gets its own isolated environment on Azure — critical for confidential deals, legal and compliance.
Keyword precision + semantic meaning. Ask in English, find Portuguese content — it understands across languages.
An internal Q&A for your team, plus client-facing bots scoped to one client's processes via magic link.
Sort by the date a document was actually issued — not when it was filed. Audit whether the correct sequence was followed.
Every process gets a summary that stays current — with optional compliance analysis flagging bribery, fraud or sanctions.
Granular permissions per business area, enforced in code — not just hidden in the UI.
Email, WhatsApp, documents and integrations — captured automatically into one private, searchable memory.
Mandu'a was built privacy-first for confidential deals, legal and compliance work. The LLM infrastructure runs on Azure with per-client data isolation — and the AI is deliberately designed never to act on its own.
Models served via vLLM/TGI on Azure infrastructure, with each client's data isolated from every other client.
The AI only ever produces text — it never takes actions. A deliberate safeguard against prompt injection.
Granular permissions per business area, enforced in code — so people only ever see what they should.
Frequently asked questions
Mandu'a is an AI application that captures emails, WhatsApp messages, documents and system data, organizes everything by process and lets you query it in natural language. The LLM infrastructure runs on Azure, with per-client data isolation.
The LLM infrastructure runs on Azure with per-client data isolation, so one company's documents never mix with another's. Access is role-based and the AI only produces text — it never takes actions.
Yes. You ask the way you'd ask a colleague and it answers citing the source documents, finding information by meaning rather than just exact keywords.
No. Mandu'a files everything automatically by process, with timelines and structured numbering. Just CC the assistant or point it at the watched folders.
Document- and deal-heavy operations: trading, import/export, logistics, accounting, legal and hospitals. Any business that must keep, prove and quickly retrieve a long paper trail.
Your Company's Knowledge Layer
Tell us a bit about your operation and we'll show you Mandu'a running on your own kind of data. We reply to every message personally.