Cited Knowledge Workspace
Knowledge & Document Software · Knowledge Retrieval / Cited Answers
Answers grounded in an organization's own documents, with every source cited
An organization needed AI over sensitive documents it could not send to an outside service. Datics is building a self-hosted workspace that ingests PDF, Word and Excel files, extracts people, places and keywords, answers questions with citations to the exact passage, and filters every answer by what the user is entitled to see.
- Status
- In development
- Engagement
- 2026
Knowledge Retrieval / Cited Answers
Ask
Where questions are answered with citations.
- Extraction
- Cited answers
- Entitlement filtering
- Self-hosted
Role- and entitlement-controlled access inside a self-hosted boundary.
The starting point
An early version existed, but its architecture overlapped and depended on a hosted backend the organization did not want to rely on.
What had to change
Processing, retrieval and writing had to work together, portably, inside the organization's own boundary.
What couldn't break
Document permissions had to carry through into every answer, conversations had to stay separate between tenants, and the model had to run on infrastructure the organization controls.
Why this was hard
Cited answers are only trustworthy if extraction quality, entitlement filtering, citation alignment and a self-hosted model all line up.
What Datics changed
Datics rebuilt it as a portable stack: role-based access for admins, supervisors, indexers and viewers; extraction with OCR fallback and quality checks; entity extraction; retrieval with citations aligned to the grounding passage; entitlement-based filtering; a writing assistant with grammar and structure checks; and a self-hosted model.
Packaged with Docker Compose for self-hosted deployment, delivered in phases against a fixed scope.
Extraction
PDF, Word and Excel ingested, with OCR fallback and quality checks.
Cited answers
Every answer links to the passage it came from.
Entitlement filtering
Answers limited to what each user may see.
Self-hosted
Runs, model included, inside the organization's own boundary.
How it works
Connects to: The organization's document collections and archives.
- 1
Ingest
Where documents are extracted, indexed and organized.
- 2
Ask
Where questions are answered with citations.
- 3
Write
Where people draft with grammar and structure checks.
Runs within: Role- and entitlement-controlled access inside a self-hosted boundary.
Engagement at a glance
Extraction, cited answers and writing run over the organization's own documents, inside infrastructure it controls.
- Deployment
- Self-hosted, model included
- Access
- Four roles with entitlement filtering
- Status
- In development
Under the hood
- Datics role
- Business analysis, full-stack and AI engineering, QA and DevOps.
- Integrations
- The organization's document collections and archives.
- Deployment
- Self-hosted, containerized deployment.
- Production boundary
- Role- and entitlement-controlled access inside a self-hosted boundary.
Technology
What this unlocked
The same cited retrieval powers Docently, which answers learner, instructor and admin questions from a product's help center.
Related work
Learning-agency directory
EdTech · Directory / Marketplace
A vetted, subscription directory that helps education buyers find and compare learning agencies
A members-only directory of vetted education agencies with search, filters, shortlists, moderated comments and Stripe subscriptions, built end to end in six milestones.
Leadership-development practice app
EdTech · Learning App / Mobile & Desktop
A leadership-development practice app taken from the web to iOS, Android and desktop
A daily leadership practice built on seven inner archetypes, taken from a web app to native iOS, Android, Windows, macOS and Linux apps from one codebase.
Voice Roleplay Trainer
Insurance Sales Training · Skills Practice / Voice AI
AI voice roleplay that makes skills practice measurable
Agents pick a scenario, a prospect personality and a difficulty, hold a live voice call with an AI prospect, and get a scorecard, transcript and coaching against the firm's own method.
Have a product with the same kind of constraints?
Tell us what already exists and what you want it to do. We will tell you what it takes.