1. Conduid
  2. RAG
  3. Veritas Vault
MCP server · RAG

Veritas Vault

Local-first AI knowledge retention engine. Passively captures sessions from Antigravity and other AI tools, stores them in SQLite with RAG, and provides morning briefs, action items, and searchable memory.

34Low

Scored 5 months ago · breakdown

About Veritas Vault

Veritas Vault is an MCP server in the RAG category: local-first AI knowledge retention engine. Passively captures sessions from Antigravity and other AI tools, stores them in SQLite with RAG, and provides morning briefs, action items, and searchable memory.

Install

From source
git clone https://github.com/VrtxOmega/veritas-vault

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

✦

Ask AI about Veritas Vault

Powered by Claude · Grounded in docs

I know everything about Veritas Vault. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

  • ✓README presentThe repository ships documentation.

5 checks have not been run yet.

README

VERITAS VAULT v3.0

High-assurance AI knowledge engine that auto-captures, organizes, and intelligences your developer work across Antigravity, ChatGPT, and Gemini sessions.


Architecture

┌─────────────────────────────────────────────────────────┐
│                    ELECTRON MAIN PROCESS                │
│                                                         │
│  main.js ─── Orchestrator (791L)                        │
│  ├── db.js ─── SQLite CAS + MiniSearch (332L)           │
│  ├── archivist.js ─── Ingest Pipeline (297L)            │
│  ├── journal.js ─── Digest Engine (~2400L)              │
│  ├── vault-ai.js ─── 19 AI Ops via Ollama (829L)       │
│  ├── vault-rag.js ─── RAG Chat Engine (575L)            │
│  ├── capture-server.js ─── HTTP Relay (292L)            │
│  ├── followup-engine.js ─── Smart Alerts (305L)         │
│  ├── watcher.js ─── FS Watcher (81L)                    │
│  ├── clipboard-monitor.js ─── Polling (68L)             │
│  └── vault-logger.js ─── Structured Logging (~100L)     │
│                                                         │
│  preload.js ─── Secure IPC Bridge (55+ methods)         │
│                                                         │
├─────────────────────────────────────────────────────────┤
│                   RENDERER PROCESS                      │
│                                                         │
│  renderer/index.html ─── 7 Panels + 4 Modals (343L)    │
│  renderer/app.js ─── Full UI Logic (~2500L)             │
│  renderer/styles/vault.css ─── Design System (71KB)     │
│                                                         │
├─────────────────────────────────────────────────────────┤
│                  BROWSER EXTENSION                      │
│                                                         │
│  extension/content.js ─── DOM Scraper (14KB)            │
│  extension/background.js ─── HTTP Relay (9KB)           │
│  extension/popup.html/js ─── Config UI (13KB)           │
└─────────────────────────────────────────────────────────┘

Features

Core Engine

  • SQLite Content-Addressed Storage — WASM-based (sql.js), atomic write-temp-rename for crash safety
  • MiniSearch Full-Text Index — Sub-millisecond fuzzy search with prefix matching
  • SHA-256 Dedup — Content hashing on all ingested files, captures, and clipboard
  • Incremental Corpus Scan — Only processes files modified since last scan
  • 7-Day Auto-Backup — Daily timestamped DB backups with automatic rotation

AI Intelligence (Ollama qwen2.5:7b)

  • 19 AI functions: summarize, classify, quality score, cluster, prioritize, enrich, extract actions
  • Keyword/heuristic fallback when Ollama offline
  • Learning loop with accept/reject feedback bias adjustment
  • Auto-ingest pipeline: background FIFO queue for non-blocking capture processing
  • Intelligence Sweep: 4-phase auto-orchestrator (re-summarize → enrich KIs → extract actions → auto-link)

RAG Chat

  • Dual-mode: semantic (nomic-embed-text 768d vectors) + keyword fallback
  • Crash-resilient indexing with checkpoint/resume
  • Memory-efficient Float32Array embedding storage
  • Cosine similarity retrieval → Top-8 chunks → Context injection → qwen2.5:7b synthesis

Capture Pipeline

  • Chrome extension captures ChatGPT, Gemini, and Claude sessions
  • Local HTTP server on 127.0.0.1:47831 with bearer token auth
  • Rate limiting (30 req/min) + CORS restricted to chrome-extension:// origins
  • Clipboard monitor: 2s polling with SHA-256 dedup and rate limiting
  • Filesystem watcher: chokidar deep watch with 10-level depth

Smart Follow-Ups

  • 5 proactive detectors: stale audits, abandoned tasks, unlinked captures, stale KIs, continuity gaps
  • Dismissal persistence via SQLite
  • Priority sorting (HIGH → MEDIUM → LOW)

Journal Engine

  • Daily digest builder with session aggregation
  • AI-powered morning brief with resume context and action items
  • Activity momentum heatmap
  • Pinned sessions with notes
  • Quick notes per date
  • Tamper-evident audit chain (SHA-256 hash chain)

UI (7 Panels)

  • Journal — Morning brief, momentum heatmap, pinned sessions, daily timeline
  • Search — Full-text fuzzy search with highlighted snippets
  • Threads — Knowledge Item threads with KI Health Dashboard
  • Tasks — Open task.md tracking with finalize/archive
  • Intelligence — AI dashboard with auto-organize, clusters, priorities
  • Ask — RAG chat with source citations and suggested queries
  • Settings — Profile, capture config, AI model info, keyboard shortcuts

Stack

Component Technology
Runtime Electron 34
Database sql.js (SQLite WASM)
AI Ollama (qwen2.5:7b + nomic-embed-text)
Search MiniSearch 7.x
FS Watch Chokidar 4.x
Markdown Marked 17.x + Highlight.js 11.x

Security

  • Context isolation + no node integration in renderer
  • CSP: default-src 'self'; script-src 'self'
  • 32-byte random bearer token auth on capture server
  • Localhost-only server binding
  • Atomic DB writes (temp → rename)
  • Single instance lock
  • 5MB request body limit

Installation

# Clone
git clone https://github.com/VrtxOmega/veritas-vault.git
cd veritas-vault

# Install dependencies
npm install

# Ensure Ollama is running with qwen2.5
ollama pull qwen2.5:7b

# Start the vault
npm start

Usage

Veritas Vault runs as an Electron desktop application. Once started:

  1. Dashboard — View session summaries, knowledge entries, and daily briefs
  2. RAG Chat — Query your captured knowledge with semantic search
  3. Journal — Browse auto-generated digest entries
  4. Watcher — Monitors filesystem changes and captures context
  5. Clipboard Monitor — Auto-captures copied text for knowledge retention

The vault passively captures your work sessions, indexes them with MiniSearch FTS and Ollama embeddings, and generates morning briefs with actionable intelligence.

📖 Read the master narrative: Why Sovereign AI?

🌐 VERITAS Omega Ecosystem

This project is part of the VERITAS Omega Universe — a sovereign AI infrastructure stack.

License

BSL-1.1

README copy captured when this server was indexed; it may lag the repository. The original is authoritative.

Questions

About Veritas Vault

How do I install Veritas Vault?

No package has been published for this server, so install it from source: clone the repository and follow its README.

Is Veritas Vault safe to use with an AI agent?

Its trust score is 34 out of 100 (low). It passes 1 of the 1 static security checks run so far. It has no ConduID identity yet, so agent calls to it are not receipted.

Is Veritas Vault still maintained?

Conduid hasn't recorded a commit date for this repository yet. Check the repository directly for recent activity.