Valora Documentation
Technical architecture, core concepts, and implementation details for the Valora AI Workflow Command Center.
The Architecture of Focus
The modern knowledge worker's workflow is fractured. Traditional email clients are passive, simply displaying chronological lists of text. Conversely, modern AI assistants lack native context; forcing users to constantly copy-paste emails to get help.
Valora solves this by acting as an embedded Chief of Staff. By connecting directly to your Google Workspace via the Corsair SDK, Valora doesn't just read your email—it actively manages it. It classifies incoming emails, detects sensitive information, extracts calendar invites natively, and maintains a long-term vector memory of your preferences.
Core Systems
Autonomous AI Priority Engine
Every incoming email is parsed through Groq (LLaMA-3.3-70b-versatile). The AI analyzes sender, context, and tone to assign a Priority Score and label (Urgent, High, Normal, Low).
Persistent Agent Memory
Valora uses the Mem0 API combined with pgvector embeddings to give your AI agent permanent memory. It remembers your preferences and applies them to future interactions.
Calendar Intelligence
Scans threads for temporal context. With a single click, Valora extracts attendees, parses natural language times into ISO formats, and generates Google Calendar invites natively.
Data Sovereignty & Security
Tokens are AES-encrypted at rest. The 'Danger Zone' utilizes safe, atomic database transactions to sever foreign-key constraints and irreversibly wipe all local data when requested.
Technology Stack
- • Next.js 15 (App Router)
- • React 19
- • TailwindCSS
- • Framer Motion
- • Zustand State Management
- • Node.js & tRPC
- • Prisma ORM v6.19
- • NextAuth (Auth.js v5)
- • Corsair Google SDK
- • Neon Serverless PostgreSQL
- • pgvector Embeddings
- • Groq API (LLaMA 3.3 70B)
- • Mem0 Memory API
Architecture Deep Dive
1. Ingestion Layer: We use Corsair's headless Google API abstraction to ingest webhooks instantaneously. Unlike legacy polling systems, Valora knows about a new email the millisecond Google receives it.
2. Inference & Classification: Raw payloads are sanitized and sent to our Groq API endpoints. Using LLaMA-3.3-70b-versatile, the system achieves 800+ tokens/second inference to classify the urgency and intent of the email before it ever renders on the client.
3. Vector Storage: Processed emails and calendar events are converted into high-dimensional vector embeddings and stored in Neon PostgreSQL using the pgvector extension. This enables true semantic search (e.g., searching "when are we deploying" actually finds the email about "release scheduled for Friday").
4. Real-time Client Sync: The frontend bypasses traditional React render cycles for heavy lists by heavily leveraging Zustand and tRPC subscriptions, ensuring sub-200ms UI interactions.
Future Roadmap: The Unified Workspace
Google Drive Integration
Seamlessly attach, search, and semantically query all your Google Drive documents directly from the command center.
GitHub Repositories
Manage pull requests, issues, and CI/CD alerts without leaving your workflow. The AI will summarize diffs automatically.
Slack & Discord
Unified messaging. Let the AI triage your Slack DMs and prioritize them alongside your urgent emails.
WhatsApp Business
Bring client communications from WhatsApp into Valora. Have the AI draft professional responses based on your email history.
REST API Reference
Local Setup
cd valora
pnpm install
# Configure your .env file with API keys
pnpm dlx prisma generate
pnpm dlx prisma db push
pnpm run dev
