abraxas914/VESTI

What it solves

Vesti addresses the problem of "data feudalism" in the AI era, where conversations with various LLMs (ChatGPT, Claude, Gemini, etc.) are locked within separate platform silos. Users often lose the trajectory of their thoughts because their data is scattered and controlled by the platform providers. Vesti provides a local-first hub to capture, consolidate, and manage these interactions, returning data sovereignty and the ability to trace intellectual evolution to the user.

How it works

Vesti operates as a two-layer system:

1. Capture Engine (Chrome Extension): A browser extension that silently monitors and extracts conversations from supported AI platforms. It uses a Parser/Observer/Gate architecture to capture dialogue content and store it in a local IndexedDB via Dexie.js. It also generates structured summaries of threads using the ModelScope API.
2. Knowledge Management Layer (Web Dashboard): An independent web view that interacts with the local data via a StorageApi. It features a Library for categorized management, an Explore module for RAG-based natural language querying of one's own history, a Network view for visualizing semantic similarities between conversations using a force-directed graph (ECharts), and a Notes section for creating curated Markdown notes.

An integrated Gardener Agent runs asynchronously in the background to classify conversations and assign semantic tags, turning raw logs into structured knowledge.

Who it’s for

Power users of multiple AI platforms who want to maintain a permanent, searchable, and interconnected record of their thinking process and avoid relying on platform-specific history logs.

Highlights

• Local-First Architecture: All data is stored on the user's device in IndexedDB, ensuring complete privacy and data ownership.
• Cross-Platform Consolidation: Automatically captures interactions from ChatGPT, Claude, Gemini, DeepSeek, Qwen, and Doubao.
• Local RAG: Implements a retrieval-augmented generation system using ModelScope embeddings and cosine similarity to answer questions based strictly on the user's own conversation history.
• Thought Topology: Visualizes the relationship between different conversations through a semantic similarity matrix and a dynamic knowledge graph.
• Structured Insights: Uses a "Prompt-as-Code" strategy to generate summaries that reveal the evolution of thought and key insights rather than just compressing text.