Holie is a self-hosted, multi-tenant AI assistant for organisations. It answers with citations from company knowledge, runs task agents, and keeps every tenant's data, access and spend isolated. It is already live in production at Atlas and at a 400-person software company.
See how Holie works → Take the product tourScreens from the production app. Names and history are sample data, and the logo area is left blank because every tenant brands Holie with its own logo and name.
Holie searches before it answers and shows every query and source it used. Tables, headings and next steps come back formatted and ready to paste into a document.
GPT, Gemini and Claude sit behind one interface, each with a plain-language hint about what it is good at. Admins decide which roles see which models.
Upload files or connect SharePoint, toggle web search, start a multi-round deep research run, or call business tools such as ClickUp and Zoho CRM.
Admins publish task agents (HR policy, newsletters, coding, CV writing) and staff start them in one click.
A project groups chats, files and instructions around one piece of work, and can be shared with colleagues.
Images and artifacts generated in any conversation land in a gallery where they can be filtered, downloaded and shared.
Policies, regulations, invoices and CVs sit in PDFs, SharePoint and OneDrive, where public chatbots can't see them.
Off-the-shelf AI tools send sensitive company and client data to third-party clouds. Many organisations can't accept that.
Per-user pricing grows with headcount, and a single-model vendor is a lock-in risk as models change every few months.
General-purpose AI doesn't know industry rules, internal templates or how a team actually works.
Holie started as client work: self-hosted assistants for Atlas (architecture and construction) and for the staff of a 400-person software company. Both now run on the same multi-tenant platform, which ships as Holie.
Teams upload files or connect SharePoint and OneDrive. Every answer cites the page it came from.
Admins configure assistants for each team. Every request is routed to the right agent and the right tools.
The research agent searches the web over several rounds and writes a structured report with sources.
Each customer is an isolated tenant. Admins decide who uses which model and how much they can spend.
Holie sits alongside two products from the same team. They share Holie's engineering team, its Microsoft 365 stack and its self-hosting model, and Holie already hands users over to Meeting Agent with single sign-on.
Every meeting, documented, and nothing leaves your tenant. The agent joins the Teams meetings it is invited to and sends structured minutes to every attendee within 15 minutes of the call ending.
DrParser turns unstructured files into reliable, structured data: it parses, extracts, splits and compares documents. It began as a CV-parsing tool for Bravo, in English and Vietnamese, and is applied in document work for clients such as Plenti and Excosoft.
Accurate parsing, extraction and comparison at scale, with minimal manual effort. It runs entirely inside the customer's environment.
Built for complex manuals and technical documents. DrParser detects tables, footnotes and layout sections that other systems often miss.
Results as published by the DrParser team; not independently verified.
Holie builds on twelve published client projects that the team behind it delivered in seven countries, from generative AI assistants to computer vision on fish farms. Figures come from each published case study.
"Atlas Assistant" lets staff chat with architecture and construction regulations stored in SharePoint (EU, Australasia and global). A Fee Proposal Agent drafts standards-aligned project proposals.
A company-wide assistant for content, code and email, plus an agent that answers HR and IT questions over internal documents while keeping data confidential.
A custom deep-learning classifier for financial documents (invoices, payslips, bank statements) and entity extraction from car invoices, so loan details can be verified automatically.
Computer vision, OCR and layout and table detection turn PDF technical documents into Excosoft's Flex format with minimal manual work.
A team of three AI engineers built DrParser to read, classify and extract data from English and Vietnamese CVs, with secure storage and smart candidate search.
A cloud, multi-tenant computer-vision system on Azure for fish counting, stress and mortality detection and wound detection, with real-time alerts.
Three products: 7-segment digital display recognition, breathalyzer digit recognition and fall detection for elderly people, running online and offline.
A fine-tuned DistilBERT with contrastive learning classifies the tone of user reviews across languages on an e-learning platform.
AI agents analyse prospect profiles through the LinkedIn API and write personalized emails at scale, integrated with OpenAI and Mailjet.
LLM-based matching connects startups with experts willing to work for equity, on an Azure cloud platform.
A modern data platform on Azure, Snowflake and Synapse with ML forecasting, anomaly-based fraud detection and real-time streaming.
A five-person data and BI team extended Salesforce reporting with Azure Synapse, Databricks and Power BI. That work grew into a global data platform.
Holie.ai takes an assistant first built for clients and turns it into a product of its own.
A dedicated Data & AI team starts in Da Nang, delivering ML projects for international clients.
ML, NLP, computer vision and data platforms for clients in Australia, Norway, Sweden, Japan, Germany, the UK and Vietnam.
A multi-tenant assistant platform is built for Atlas and a 400-person software company, and goes live for both.
The assistant gets its own brand and home at holie.ai.
See Holie running on your own documents, inside your own environment.
contact@holie.ai → Product tour