Skip to content

update_v0.3.md

Latest commit

 

History

History
92 lines (75 loc) · 7.87 KB

update_v0.3.md

File metadata and controls

92 lines (75 loc) · 7.87 KB

v0.3 Series: Identity, Structure & QA

v0.3.0: Identity & Structural Mesh

Date: December 29, 2025

👤 Identity & Persona System

  • Identity View: Introduced a completely new interface for managing User and AI identities.
    • Dual Profiles: Create distinct profiles for the "User" (Observer) and the "AI" (Agent).
    • Dynamic Switching: Swap between agent personalities (e.g., "Analyst", "Creative", "Skeptic") on the fly to change the tone and perspective of the synthesis.
    • Contextual Injection: Selected identity traits are now dynamically injected into the system prompt, influencing every generation.

🗄️ Dataset Management (formerly Sessions)

  • Multi-Dataset Architecture: Replaced the ephemeral "Sessions" model with persistent Datasets.
    • Isolation: Each dataset acts as an independent Knowledge Graph container.
    • Switching: Seamless toggling between different projects or knowledge bases via the sidebar.
  • Import/Export: Added capability to import external datasets (JSON) to hydrate the graph instantly.
  • Database Service: Refactored dbService.ts to handle isolated graph persistence and retrieval.

🕸️ Structural Intelligence (Auto-Connect Mesh)

  • Mesh Topology: Overhauled the "Auto-Connect" logic to generate Full Mesh structures for categorical data.
    • Tables as Clusters: Instead of a "Star" topology (root -> everything), the engine now creates highly interconnected "Table Clusters" (all-to-all).
    • Graph Sanitization: Implemented rigorous input sanitization to prevent D3 object mutation bugs that previously caused "invisible links".
  • Optimized Weighting Strategy:
    • Backbone (0.85): Table-to-Table connections are set to high strength to maintain structural integrity.
    • Content (0.55): Item-to-Table connections start at moderate visibility (> theta 0.30), leaving 45% headroom for Hebbian Learning to distinguish "useful" nodes over time.

🧠 Prompt Engineering

  • Dynamic Prompt Engine:
    • Context-Aware: Prompts now dynamically assemble themselves based on the active Persona, User Preferences, and Current Task.
    • History Injection: Integrated ChatHistory directly into the prompt context window.

🎛️ Interface Controls

  • Context Window Slider: Added a precision slider to the Right Panel (Chat History).
    • Granularity: Dynamically adjust how many previous messages (0-10) are included in the immediate context window.
    • Effect: Allows users to "Narrow" focus (0 history) for unrelated questions or "Widen" focus (10 history) for complex multi-turn conversations.

📚 Documentation

  • Architecture Notes: Consolidated scattered research notes into a centralized architecture_notes.md.
  • Reference Cleanup: Audited and fixed all documentation links to ensure consistency across the project.

🛠️ Maintenance & Stability (Hotfix)

  • Robust Feature: Fuzzy Entity Grounding: Resolved critical retrieval failures in complex datasets (like Sci-Fi) by implementing fuzzy label matching. This allows the AI to correctly map human concepts (e.g., "Books") to system UUIDs, ensuring consistent context retrieval.
  • Crash-Proof Logging: Hardened the Trace & Safety logging pipeline. Logs are now reliably captured and displayed even if the internal processing pipeline encounters errors, providing transparent debugging.
  • Architectural Sanctity: Enforced strict separation between Data (Datasets) and Configuration (Global). Patched scifiData.ts to remove conflicting Empty API Keys, ensuring that all datasets correctly inherit your global credentials.

v0.3.1: Quality of Life & Stability Fixes

Date: December 30, 2025

🔥 Hotfix: Identity & State Integrity (v0.3.1.1)

Issue: An issue was detected where the AI would persist with default identity traits (e.g., "Jade") even after the user selected a different persona (e.g., "Emma"), due to a combination of React State caching and Ghost Nodes in the retrieval graph.

Fixes Applied:

  1. Direct State Overrides: The ChatView now explicitly enforces the locally selected identity for every query, bypassing potential global state lag.
  2. Context Sanitizer: The RAG Pipeline now aggressively scans retrieved context for conflicting Identity Preferences (e.g., "AI Name: Jade") and purges them before they reach the LLM.
  3. System Prompt Fallback: Added a safety mechanism to ensure {{char}} placeholders are always injected, even if the user's template is malformed.
  4. Dataset Clean State: Resolved a bug where creating a New Dataset would erroneously pre-fill it with the default "Demo" data. New datasets now initialize as completely empty canvases, as intended.
  5. Dangling Synapse Crash: Fixed a critical pipeline failure where the engine crashed when processing synapses connected to deleted nodes. Added safety checks in the Orthogonality enforcement logic.

🚀 Key Improvements

1. Active Focus & Context Logic

Major improvements to how the "Working Memory" (Active Focus) behaves, giving users precise control over what is currently active in the Context Engine.

  • Unified Interface: Merged "Active" and "Available" pools into a single, clean list of chips.
  • Explicit Deletion ("X" Button): The 'X' button on chips now permanently deletes the node from the dataset (after confirmation), ensuring you can curate your graph effectively.
  • Toggle Behavior: Clicking the chip body now correctly toggles the Active/Inactive state in working memory without deleting the data.
  • 3-Item Limit: Implemented a strict limit of 3 active context items to ensure focused analysis. Users must deselect an item to add a new one if the limit is reached.
  • "Restore" Button: Renamed "Clear All" to "Restore". This button now resets the Graph Nodes and Active Selection to their default state for the dataset, but preserves your Chat History.

2. Application Stability & Navigation

  • Factory Reset ("Reset All"): Added a red "Reset All" button in the App Header. This performs a complete Factory Reset of the current dataset, clearing:
    • Graph & Synapses
    • Chat History
    • Telemetry & Trace Data
    • Audit & Debug Logs
    • Safety Rule Status (Broken Rules)
  • Rendering Fixes: Resolved critical rendering crashes related to function nesting in the main application logic.

3. Lattice View

  • Auto-Connect Consistency: The "Auto-Connect" button in Lattice View now uses the exact same logic as the initial dataset loader, ensuring consistent mesh generation for manually added nodes.

🔧 Hotfix 2: Type Hardening & Component Wiring (v0.3.1.2)

Issue: A series of TypeScript compilation errors and a component linkage failure were checking critical paths in the app. Fixes Applied:

  1. Duplicate Identifiers: Resolved a Duplicate identifier error in AppHeader caused by conflicting prop definitions.
  2. Missing Component Wiring: The GraphExplorer was attempting to import a non-existent GraphVisualizer. It has been correctly repointed to VisualGraphView.
  3. Strict Type Compliance: Fixed AuditLog type mismatches in EvalRunner and removed unsupported generic type arguments from D3 calls in VisualGraphView.
  4. Pipeline Stability: Hardened useChatPipeline.ts against a "Property 'id' does not exist" crash during context sanitization.
  5. About Page: Updated personal links and added functional social share buttons for Twitter, LinkedIn, Facebook, and Reddit.

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.