Analyze operational documents without sending them to the cloud.
The native Documents workspace imports local reference material, searches it, answers grounded questions, and produces cited summaries while disconnected—keeping file content inside the application process.
Windows x64
Supported platform or deployment profile.
Android
Supported platform or deployment profile.
Offline
Supported platform or deployment profile.
Memory-only index
Supported platform or deployment profile.
Source citations
Supported platform or deployment profile.
Turn a field reference library into a private, cited assistant.
Users can load multiple supported documents, retrieve the most relevant passages, ask questions, and summarize source material without requiring a network connection or cloud-model account.
Multi-document import
01
Load EPUB, text-based PDF, Open XML DOCX, Markdown, and plain-text files through the native file picker.
Local extraction and indexing
02
Extract, normalize, chunk, and index document text within bounded device-memory and archive-safety limits.
Cited search
03
Rank relevant passages locally and return the originating PDF page, EPUB chapter, DOCX heading, Markdown heading, or text chunk.
Grounded questions and answers
04
Answer from retrieved source passages and display citations so the operator can verify the underlying material.
Offline summaries
05
Create concise source-grounded document summaries using the built-in extractive engine or the approved local Windows model.
Local Phi-4 generation
06
On Windows x64, optionally run Microsoft Phi-4 Mini Instruct in process through ONNX Runtime GenAI when approved model weights are installed.
Automatic fallback
07
If the local model is absent, corrupt, out of memory, cancelled, or unable to generate, return a cited extractive result instead of losing the workflow.
Private runtime boundary
08
Keep file bytes, extracted text, retrieval, and analysis inside the app; the document-intelligence service contains no cloud-model network dependency.
Every answer leads back to the source.
The analysis pipeline retrieves a small set of relevant passages, treats them as untrusted source content, constrains the response to that evidence, and returns the corresponding citations.
- Local extraction and BM25-style passage ranking
- Page, chapter, heading, section, or chunk citations
- Retrieved excerpts only—not an unrestricted document prompt
- Model-independent extractive answer and summary path
- User verification before operational or clinical use
Useful AI where connectivity and data sensitivity matter.
The document workspace is designed for controlled devices, intermittent networks, operational references, and deployments that cannot assume cloud-model access.
In process
Memory only
No cloud fallback
Bounded input
One document workflow, with platform-specific inference.
The complete local extraction, search, cited question-answering, and summarization workflow runs on Windows x64 and Android. The optional generative provider is currently a Windows capability.
- Windows x64: extractive analysis plus optional local Phi-4 Mini generation
- Android: complete extractive search, Q&A, and summaries
- Model absent: the workspace remains operational through extractive fallback
- Text-based PDFs: indexed with page citations
- Scanned PDFs: require a separately approved offline OCR component
Provision models deliberately.
DEPLOYMENT PROFILE
The approximately 4.93 GB Windows INT4 model is installed once during deployment or placed in an authority-managed model location. It is not bundled silently, and the app verifies the required model files before enabling generation.
A governed service layer for local and connected intelligence.
Document intelligence is the completed offline baseline. Additional AI services are enabled only with the required model, device support, data policy, validation, and explicit deployment status.
Document intelligence
RELEASED / OFFLINE
Local extraction, cited retrieval, grounded questions, summaries, and resilient fallback on Windows and Android.
On-device language model
RELEASED / WINDOWS
In-process Phi-4 Mini generation for retrieved document excerpts when approved weights are installed.
Voice and content assistance
CONFIGURED MODULE
Voice-directed field tools, rich-content assistance, and language tasks are scoped against the selected device and approved model provider.
Geospatial intelligence
CONFIGURED MODULE
Field, mapping, proximity, route, and location-enrichment services require customer data, accepted rules, and deployment validation.
Vision and media analysis
EXPERIMENTAL / VALIDATE
Object, scene, OCR, image, and video analysis modules remain separate from the released document-intelligence baseline and require validated models.
External AI providers
OPTIONAL / CONNECTED
Cloud or enterprise AI services can be configured behind an explicit network, privacy, provenance, failure, and data-egress boundary.
Analysis supports the operator; it does not replace professional judgment.
On-device AI outputs are decision-support material. Deployment policy defines approved content, intended use, prohibited use, review, logging, model management, and escalation.
Check the citation
VERIFY
Operators can inspect the page, chapter, heading, or passage before using an answer.
No autonomous clinical action
BOUND
The feature does not diagnose, prescribe, dispatch, or replace clinicians, policy, medical direction, or local authority.
Manage models and content
GOVERN
Approve model versions, document sets, device profiles, updates, retention, and acceptable use for each deployment.
Bring private analysis into the operational workspace.
Connect offline intelligence with documents, field operations, configurable forms, and separately governed decision support.
Documents & media
CONTENT
View operational content, author rich text, and work across multi-view layouts.
Native field operations
FIELD
Use private analysis alongside offline mapping and response workflows.
Configurable forms
FORMS
Collect structured field data under customer-defined validation and governance.
Decision support
SUPPORT
See the separate, provenance-labelled dispatch and ETA assistance pathway.
Bring the documents, devices, models, security policy, and offline conditions that matter.
We will map supported formats, model provisioning, platform behavior, citations, content governance, performance, and acceptance tests.