Desktop app product overview¶
This page is the product-level description of the VERA desktop app (moved from the repository root). For install and first-run steps see Getting started; for internals see Architecture.
What This App Is¶
This application is a grounded document assistant built around the .vera format.
It helps users search long source documents, get useful answers, and verify those answers against the original pages.
In short: it turns document Q&A into a transparent, source-backed workflow.
Who It Is For¶
- Teams working with large manuals, policies, standards, or reports
- Analysts and compliance users who need citation-ready answers
- Engineers and operators who need fast lookup in technical documents
- AI-assisted workflows that require grounded, auditable outputs
Core Value¶
- Better search over long documents
- Answers tied back to source pages and sections
- Fewer hallucinated responses through grounding and citation
- One reusable
.verafile that works across tools and sessions
Key Capabilities¶
1. Source Document Viewer¶
- View original document pages
- Navigate by page and section
- Jump from answer citations directly to source location
- Work in a two-pane layout with Ask on the left and Source Document on the right
- Open documents from the native File menu and show file metrics in the bottom status bar
- Drag the Source Document divider to resize the grounded PDF review area
2. Prompt Input + Retrieval¶
- Accept natural language user prompts
- Retrieve relevant chunks from
.verausing keyword, semantic, or hybrid search - Pass retrieved context into response generation
3. Visual Grounding¶
- Highlight retrieved passages in the source view
- Show where each claim came from (page and heading path)
- Keep selected citations focused on the source PDF, with metadata and retrieval details available on demand
4. Session Management¶
- Save conversations and retrieval state
- Revisit prior prompts, results, and citations
- Support iterative research and comparison across runs
5. Configurable Instructions¶
- Layered instructions for response behavior
- Configurable augmentation that combines system/app instructions, retrieved context, and user prompt
- Optional domain-specific response templates
6. LLM Ask¶
- Connect to one or more LLM providers under File > Settings
- Select model by task profile (speed, quality, cost)
- Stream grounded answers with citation links
- Search remains fully local when no provider is configured
7. External Tool Connectivity¶
- Integrate useful supporting tools (search, APIs, data sources, utilities)
- Use tool outputs as additional context
- Keep provenance so users can see what informed the answer
How It Works (High Level)¶
- User asks a question
- App retrieves relevant context from
.vera - App composes prompt with instructions + context + user input
- App returns a grounded cited answer from the configured LLM provider, or Search-only results when no provider is set
- Citation clicks open the Source Document viewer and highlight the supporting region
Why .vera Matters in This App¶
A .vera file is a portable retrieval archive that packages:
- source document content
- structured text blocks and chunks
- keyword index
- embeddings
- citation metadata
- visual grounding regions
This lets the app deliver faster, more consistent, and more explainable answers than querying raw PDFs alone.
Design Principles¶
- Grounded first: answers should be traceable to sources
- Transparent by default: show evidence and retrieval path
- Configurable behavior: adapt to team and domain needs
- Tool-agnostic architecture: integrate models and utilities safely
- Reproducible sessions: preserve context and configuration history
Example Use Cases¶
- Compliance question answering with page-level citations
- Technical operations lookup in large manuals
- Policy interpretation with source traceability
- Analyst research workflows with persistent sessions
Success Criteria¶
- Users can find relevant answers quickly
- Answers include clear evidence paths
- Teams trust outputs because sources are visible
- Configuration and session history support repeatable workflows
Future Enhancements¶
Libraries (folder-scoped Search/Ask with persistent collection indexes) and LLM Ask are already shipped. Remaining product follow-ups:
- Voyage and Ollama embeddings after an optional query/document hint on
EmbeddingFunction. OpenAI already ships asvera-embed-openai. - Advanced reranking and confidence scoring
- Evaluation dashboard for groundedness and retrieval quality
- Role-based governance and audit trails
- Stronger tool orchestration and approval policies