Web app
Use AIDocumentIndexer from any modern browser, served from your own infrastructure.
AIDocumentIndexer · Private document AI
Your documents, answered. On your own servers.
AIDocumentIndexer turns contracts, reports and slide decks into a private knowledge base. Ask in plain language, get answers with citations to the exact passage – self-hosted, works with local models, optionally fully offline.
At a glanceUpdated Oct 2026
- What it is
- A self-hosted AI assistant for your organization’s documents
- Answers
- In plain language, with a citation to the exact source passage. If nothing is found, it says so.
- Deployment
- On your own servers or cloud account; air-gapped option
- Models
- Local models, or OpenAI, Anthropic, Google Vertex, AWS Bedrock, OpenRouter
- Sources
- 30+ file formats · 40+ connectors
- Access
- Web app · desktop app · browser extension
- Made by
- Mandala Labs GmbH, Berlin
The interface
Ask a question. Check the answer.
AIDocumentIndexer is a private AI assistant for your organization’s documents: self-hosted, plain-language, every answer cited. Each claim links to the passage it came from, so your team can verify instead of trust.
How it works
From your files to a cited answer
Knowledge is usually scattered across drives, inboxes and PDFs, and search only works if you know the file name. AIDocumentIndexer puts it behind one question box.
Connect your sources
Link 40+ connectors such as Google Drive, SharePoint, Slack or Confluence, or upload files. After the first sync, only changes are reloaded.
Index
Documents in 30+ formats, including scanned images (OCR) and web pages, become a searchable knowledge base on your own infrastructure.
Ask
Type a question in plain language. Pick a local model or a cloud provider; switching is a settings change.
Get a cited answer
Every answer points to the exact source passage, with the strength of evidence shown per claim. If nothing relevant is found, it says so instead of guessing.
Features
More than a search box
Answers are the core. Around them sit the tools an organization needs to work with its documents day to day.
Cited answers
Every answer points to the exact source passage, colour-coded by how well it is supported. If there is no answer, it says so.
Your sources, connected
40+ connectors from Google Drive to SharePoint; documents in 30+ formats, including scans via OCR.
Documents from your material
Generate Word, PowerPoint, PDF and Excel files from your own documents. Edit the outline first and regenerate single sections.
Knowledge graph and comparison tables
See people, organizations and concepts as a graph. Compare, say, 20 contracts across 5 criteria, with every cell cited.
Agents and workflows
Agents that run on a schedule or when new documents arrive, and visual workflows with approval steps. Every run can be replayed.
Control and audit
Access tiers, per-collection allow and deny rules, an audit log with CSV export and cost budgets.
One index, several ways in
The same index serves the web app, a desktop app and a browser extension, plus an MCP and OpenAI-compatible endpoint.
Audio overviews
Turn material into listenable summaries in six formats, for example a discussion between two speakers.
Deployment and privacy
Runs on your infrastructure
You decide where documents, index and models live. With fully local models, no document content leaves your network. If you choose a cloud model provider, queries go to that provider.
Docker · Helm
Self-hosted
Runs on your own servers or in your own cloud account. Access through the web app, the desktop app or the browser extension.
Offline-capable
Air-gapped
Local mode with local models searches without an internet connection. An air-gapped Docker configuration for isolated networks is included.
Local · OpenAI · Anthropic · Vertex · Bedrock
Local or cloud models
Use models on your own hardware, or a provider you choose. Switching is a settings change, not a migration.
Opt-in by default: anything that costs money, hides content or moves data is off until an administrator turns it on, for example sharing or video indexing. Registration is invitation-only.
- GDPR-ready
- Self-hosted
- Invitation-only registration
- Audit log with CSV export
W.A.M.
What a mess. Was a mess.
The same engine, built for the files in your own home. W.A.M. is a desktop app for private households: it runs on your own computer, answers questions about your files and puts your folders in order.
Ask your files
Question files on your PC and in Google Drive, OneDrive or Dropbox in plain language, and get answers with the source. iCloud works through its local folder.
Sort a folder with one sentence
Describe what you want. W.A.M. shows a plan with a before and after preview and changes nothing until you confirm. Every move can be undone.
Downloads that tidy themselves
New files are filed automatically. You get a notification with Undo, and W.A.M. asks when it is unsure.
Private by design
The AI is built in and works fully offline; your own cloud AI key is optional. It touches only the folders you choose, deletes nothing, overwrites nothing, and stores its data locally and encrypted.
Planned for macOS and Windows; 16 GB RAM recommended. Name and release date are not final.
Get notified at launch
One email when W.A.M. is available. Nothing else.
Who it’s for
For organizations that cannot hand their documents to a third party
If your team wants ChatGPT-style access to its own knowledge, but contracts, research and internal reports must stay under your control, this is the case AIDocumentIndexer was built for.
Teams with years of documents
Presentations, reports and contracts that nobody can search by content any more, and work that gets done twice because nobody knew it existed.
Research institutes
Large report collections that have to be searchable offline, for example for climate research in the field, with local models.
Policy teams and multi-institute cooperations
Many sources, several partners: one knowledge base where access tiers and per-collection rules decide who sees what.
These are example scenarios from our product documentation, not customer references.
Integrations
Connects to where your documents already live
40+ connectors, 30+ file formats and your choice of AI model: connect what your team already uses.
Connectors40+
- Google Drive
- SharePoint
- Slack
- Confluence
- Salesforce
- Snowflake
…and more. After the first sync, only changes are loaded.
File formats30+
- Word
- PowerPoint
- Excel
- Scanned images (OCR)
- Web pages
…and more.
AI models
- Local models
- OpenAI
- Anthropic
- Google Vertex
- AWS Bedrock
- OpenRouter
Local setups work, for example, via Ollama.
Platforms
One index, three ways in
Web app, desktop app and browser extension all work on the same index.
Desktop app
For macOS, Windows and Linux. Run it in local mode with a local AI model and offline search, or connect it to your server.
Browser extension
For Chrome and Firefox. Brings the same index into the browser you already work in.
FAQ
Questions about AIDocumentIndexer
Something missing? Write to contact@mandala-labs.com.
What is AIDocumentIndexer?
AIDocumentIndexer is a self-hosted AI assistant that indexes your organization’s documents so staff can ask questions in plain language and get answers with citations to the exact source passages.
Does my data leave my network?
Not with local AI models: the software 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.
Can it run without internet (air-gapped)?
Yes, in local mode: a local setup with local models (for example via Ollama) supports offline search, and an air-gapped Docker configuration is included.
Which file formats and sources are supported?
30+ formats including PDF, Word, PowerPoint, Excel, scanned images (OCR) and web pages, plus 40+ connectors such as Google Drive, SharePoint, Slack, Confluence, Salesforce and Snowflake.
Which AI models can I use?
OpenAI, Anthropic, Google Vertex, AWS Bedrock, OpenRouter, or models on your own hardware. Switching the model is a settings change.
How is access controlled, and is it GDPR-ready?
Access is controlled with access tiers (1 to 100), per-collection allow and deny rules, invitation-only registration and an audit log with CSV export. These controls are available for operators to use in their own GDPR compliance work; the product itself is not certified.
Is there a version for home users?
A desktop app for private households, working title W.A.M., is in development for macOS and Windows. It uses the same engine to answer questions about your own files and to sort folders, and runs on your own computer. It is not available yet.
Request a demo
See it on your own documents
We set up a demo and walk your team through it. Tell us briefly what kind of documents you work with.
- contact@mandala-labs.com
- Address
- Mandala Labs GmbH
Goßlerstr. 2, 12161 Berlin, Germany