Summary

VESTI provides a local-first AI conversation memory hub, addressing data fragmentation and ownership issues across major AI platforms like ChatGPT, Claude, and Gemini. It empowers users to capture, search, summarize, and export their AI interactions, consolidating personal dialogue history into a controllable knowledge base for enhanced self-understanding and data sovereignty.

How It Works

The system features a dual-layer architecture: a Chrome extension (Capture Engine) for background conversation capture and local IndexedDB storage, and a standalone Web Dashboard (Knowledge Management Layer) for advanced interaction. VESTI prioritizes "local-first" and "attention sovereignty," ensuring data remains on-device and users define valuable memories. Key technologies include local vector embeddings for semantic search, RAG for AI-powered Q&A, and an asynchronous "Gardener Agent" for automated summarization and classification.

Quick Start & Requirements

• Installation: Offers an offline package (unzip, load unpacked extension in Chrome developer mode) and a developer build from source (Node.js 18+, pnpm required).
• Prerequisites: Node.js (18+), pnpm (for source build).
• Configuration: A ModelScope API key is needed for summarization and vectorization; users configure this in settings.
• Links: Official releases on GitHub Releases; backup source in vesti-release/.

Highlighted Details

• Absolute Data Sovereignty: All data stored locally in IndexedDB, no cloud upload.
• Cross-Platform Capture: Supports ChatGPT, Claude, Gemini, DeepSeek, Qwen, Doubao.
• Local RAG & Semantic Search: Natural language querying of personal AI history with sourced answers.
• Knowledge Graph Visualization: Maps semantic relationships via vector similarity.
• Gardener Agent: Automates classification and tagging with observable decision-making.
• Progressive Disclosure UI: Multi-layered interface guides users from simple capture to advanced management.

Maintenance & Community

Contributions welcomed via GitHub Issues/PRs. Community channels mentioned but not linked.

Licensing & Compatibility

• License: MIT License.
• Compatibility: Permissive for commercial use, requires attribution.

Limitations & Caveats

Browser-based vector retrieval scales poorly beyond thousands of conversations. Cross-device sync is absent (deliberate local-first choice pending secure encryption). Parsers may break with AI platform UI updates. Full Weekly Digest feature is under development.