Document Q&A
Drop in PDFs, Word docs, and spreadsheets and ask plain-language questions. Every answer is cited back to the source page; query across multiple documents in one session.
Features
Everything you need. Nothing phoning home.
Purpose-built workflows for document-heavy professionals — summarize, redact, translate, transcribe, review contracts, and more — running entirely on your own machine.
Summarization
Condense long reports into bullet points, an executive brief, action items, or a full summary. Export to TXT, DOCX, or PDF — documents longer than the context window are handled automatically.
Private Writing Assistant
Draft, rewrite, shorten, and grammar-correct sensitive correspondence — formal, casual, diplomatic, or assertive — with tracked changes so you see exactly what the AI touched.
PII Redaction
Detect and remove names, addresses, IDs, emails, dates of birth, and financial accounts. Review every redaction, then export a truly redacted PDF or DOCX — the underlying text is removed, not just covered.
Private Translation
Translate documents and text across English, Chinese, Malay, Japanese, Korean, and major European languages — formatting preserved, with a side-by-side original/translation review.
Meeting Notes
Transcribe meetings, interviews, and calls from audio/video files or live mic, with speaker labels, then auto-summarize into action items and decisions.
Entity Extraction
Pull structured entities — names, organizations, dates, monetary amounts, obligations — out of unstructured documents, exportable as a CSV or XLSX table.
Contract Review
Purpose-built legal analysis: identify key clauses, flag risks by severity, summarize obligations, and detect asymmetric terms — with comparison against standard clause templates.
OCR & Document Extraction
Scanned PDFs, faxes, and photos are detected and OCR'd automatically inside any workflow — across many languages, with deskew and noise reduction for accuracy.
Advanced workflows
Scale up and automate.
Persistent knowledge, batch automation, and integration — included in the Professional and Business editions.
Local Knowledge Base Professional
Build a persistent, indexed corpus of your documents in named collections and query it across sessions with citations — no re-importing files each time.
Batch Processing Professional
Run any workflow across an entire folder, unattended — 200 contracts, 50 transcripts — with per-file progress, skip/retry, and a consolidated report.
AI Infographic Generator Professional
Turn a document or raw data into a visual infographic in a split chat-and-canvas view. Pick a style and layout; export as PNG, PDF, or SVG.
Local API Server Business
Expose an OpenAI-compatible API on localhost so your own scripts and apps can call Tholos AI — strictly on-device, with a local API key and no outbound path.
Voice & accessibility
Speak and listen — privately.
All speech processing runs on-device, unlike OS dictation and cloud voice services that send audio off your machine.
Offline Dictation
Real-time, continuous speech-to-text for composing documents, emails, and notes — with automatic punctuation and inline editing, usable across every view.
Voice Query Input
Speak a question to the chat instead of typing — single-utterance capture with automatic send, for quick hands-free Q&A.
Read Responses Aloud
Have any AI response, summary, or translation read aloud on-device, with multiple voices and 0.5–2× speed — great for hands-free review or reading fatigue.
Privacy by architecture
Not a policy. A property of the software.
There is no network path between your data and the outside world — and you can verify it.
No data leaves your device
Every operation — inference, retrieval, transcription, OCR — runs locally. The UI is configured to block all external network requests.
No account, no telemetry
No login or registration, and no analytics, crash reporting, or usage tracking of any kind. The app works fully offline after a one-time license check.
Air-gap ready
The only outbound call is an optional model-catalog check carrying zero user data — disable it and run completely disconnected.
Open and verifiable
Open-weight models under their real names, SHA-256-verified before loading, with all your data in standard, user-accessible folders.
Runs on your hardware
Models sized to your machine.
Tholos AI picks a model tier that fits your hardware on first run — you never deal with quantization formats or parameters.
Light
3–4B params
Fast summaries, simple Q&A, and entity extraction. Runs comfortably on 8 GB RAM.
Balanced
7–8B params
Strong instruction following and multilingual work. The sweet spot for most users.
Power
14B params
Highest-quality reasoning for contract review and complex analysis.
Workstation
100B+ params
Frontier-class reasoning on workstation-grade hardware. Opt-in.
Chat directly with any model you load, or run the purpose-built workflows above. Prefer your own weights? Drop in any GGUF model.
Choosing the right model for your hardware →
How it works
From install to insights in three steps.
01
Install
Single installer for Windows or macOS. No accounts, no signup, no telemetry.
02
Load models
Download an LLM, ASR, and OCR model from the in-app catalog — or drop in your own GGUF / ONNX files.
03
Work offline
Disconnect from the internet entirely. Tholos AI keeps working — chat, RAG, OCR, transcription, all of it.
See it run on your machine.
Try every feature free for 14 days — no account, no credit card.