PRODUCT_POSITIONING_AND_EVIDENCE_THREADS_AUDIT.mdHistoricalView source

Documentation status: [Archived]. This document is kept for historical evidence. Do not use as the source of truth for OpenIntelligence v4.3.

Product Positioning & Evidence Threads Audit

1. Product Truth Audit

OpenIntelligence is fundamentally different from generic “document AI” wrappers. An analysis of the architecture reveals its core truths:

  • Local-First & Privacy-Preserving: The pipeline uses CoreML and apple_fm_embed (Apple Foundation Models) for embeddings. It is designed to run inference locally, meaning sensitive documents never leave the device.
  • Extreme Transparency: The presence of PipelineTraceExporter and rich ThinkingEvent structures shows that the app doesn’t just return answers—it proves its work. It exposes retrieval times, chunk ranking, and metadata generation.
  • Ecosystem Native: Deep integration with AppShortcutsProvider (Siri), WorkspaceSyncService (iCloud), and the Apple Silicon hardware stack makes it a native tool rather than a web port.

2. Buyer Motivation Inference

Buyers in the current market are fatigued by the “chat with PDFs” trope. They are likely downloading and upgrading OpenIntelligence because:

  • Data Sovereignty: They handle sensitive IP, financial records, or legal documents that cannot legally or ethically be uploaded to third-party APIs (OpenAI, Anthropic).
  • Trust & Verification: They cannot afford hallucinations. They are buying the inspectability of the app—the ability to trace a claim back to the exact source chunk.
  • Workflow Durability: They want a tool that acts as a persistent research assistant across their Apple devices, not an ephemeral chat bot they have to re-prompt every session.

3. Anti-generic Positioning

To escape the “PDF chatbot” category, OpenIntelligence must be positioned as an Inspectable Evidence Engine.

10 Positioning Statements:

  1. Your personal intelligence engine. Private, local, and fully inspectable.
  2. An evidence engine for your documents, native to Apple Silicon.
  3. Stop chatting with PDFs. Start building verified evidence threads.
  4. The transparent research assistant that never sends your data to the cloud.
  5. Local RAG that proves its work.
  6. Your private knowledge base, supercharged with inspectable Apple Intelligence.
  7. Answers you can trust, backed by a fully auditable pipeline trace.
  8. iCloud-synced research threads with zero-knowledge data extraction.
  9. An intelligence layer for your files that respects your privacy.
  10. The Apple-native evidence engine for professionals.

Taglines:

  • Truth, extracted locally.
  • Intelligence, verified.
  • Your private evidence engine.

App Store Subtitles:

  • Private Evidence Engine
  • Local Research Assistant
  • Auditable Document RAG

4. Evidence Threads Concept Audit

Currently, the chat architecture (ChatMessage.swift and ChatScreen.swift) treats queries largely as ephemeral point-in-time interactions scoped to an active containerId. While a ChatMessage heavily stores its retrievedChunks and pipelineTrace, the overarching conversational session is fragile. If the user clears the chat or switches libraries, that train of thought is lost. There is no durable “Thread” or “Session” object to group a continuous line of investigation into a retainable, shareable asset.

5. Evidence Threads MVP Design

To pivot to a durable workflow, the app must introduce the Evidence Thread.

  • Data Model: A new SwiftData entity, EvidenceThread, which has a one-to-many relationship with ChatMessage.
  • Properties:
    • id: UUID
    • title: String (auto-generated by the LLM based on the first query)
    • containerId: UUID (tying the thread to a specific Library)
    • createdAt / updatedAt: Date
    • summary: String? (A user-triggered synthesis of the entire thread)
  • Value: This turns a fleeting Q&A session into a durable research artifact that can be revisited, appended to, or exported.

6. UX Loop

  1. Initiation: The user selects a Library and asks a question. A new EvidenceThread is silently created and titled.
  2. Investigation: The user continues asking follow-up questions within this Thread. The UI clearly visualizes the source chunks and pipeline trace for each step, reinforcing the “evidence” aspect.
  3. Persistence: The thread is automatically saved and synced via iCloud (WorkspaceSyncService). It appears in a new “History” or “Threads” sidebar.
  4. Synthesis & Export: The user can tap “Summarize Thread” to generate a cohesive brief of their findings, and export the entire thread (including source citations) via an extended PipelineTraceExporter.

7. App Intents Integration

The current RAGAppIntents.swift provides excellent zero-friction entry points (QueryDocumentsIntent, AskDocumentIntent). This workflow can be supercharged by adding:

  • ContinueEvidenceThreadIntent: Allows a user to dictate a follow-up question to a specific active thread via Siri while on the go.
  • SummarizeThreadIntent: “Hey Siri, summarize my Q3 Revenue evidence thread.” This makes the evidence gathering process ambient and continuous.

8. Landing Page Conversion

The marketing materials and App Store screenshots must stop showing generic chat bubbles.

  • Visuals: Showcase the PipelineTraceExporter UI, highlighting the transparency of the retrieval (similarity scores, chunks, paths).
  • Messaging: Lead with “See exactly where the answer came from.”
  • Privacy Badge: Prominently display a “100% On-Device Inference” badge to capture the privacy-conscious enterprise/legal/medical buyer.

9. Monetization Implications

The EntitlementStore currently limits users by document limits and library counts. Evidence Threads provide a powerful new lever for the Pro and Lifetime tiers:

  • Free Tier: Ephemeral Q&A or a strict limit of 3 active Evidence Threads.
  • Pro Tier: Unlimited Evidence Threads, iCloud syncing of threads across devices, and the ability to export full thread dossiers to Markdown/PDF. This aligns the payment trigger directly with the core value: durable, retainable research output.

10. Final Implementation Plan

  1. Model Layer Update: Create the EvidenceThread model and update ChatMessage to include a threadId foreign key.
  2. State Management: Update RAGService and ChatScreen.swift to manage an activeThreadId.
  3. UI Additions:
    • Add an “Evidence Threads” history sidebar or drawer to the UI.
    • Implement thread renaming, deletion, and summary generation.
  4. Export Pipeline: Extend PipelineTraceExporter to format an array of ChatMessages into a single cohesive document.
  5. App Intents Update: Implement intents for querying within a specific thread context.