Technical Architecture

Built with
cutting-edge AI

A production-grade agentic application powered by Claude, vector search, and modern cloud infrastructure.This is the future of personal knowledge management.

9+
AI Tools
1536
embedding dims
SSE
streaming
100%
type-safe
Agentic AI System

Not just an LLM wrapper

A full agentic system with tool calling, memory, vector search, and context awareness. Built on Anthropic's Claude with custom architecture.

Multi-Turn Conversations

Stateful chat sessions with full conversation history and context awareness across turns

Tool-Calling Architecture

Dynamic tool registry with 9+ tools: search, CRUD, counts, notifications, briefing generation

Real-Time Streaming

Server-Sent Events (SSE) for live token streaming and tool execution feedback

RAG with Vector Search

pgvector embeddings (1536-dim) for semantic search over user knowledge base

Intelligent Memory

UserMemory system stores learned preferences, context, and interaction patterns

Context-Aware Responses

Agent understands urgency, time-sensitivity, user intent, and action requirements

Advanced Features

Enterprise-grade capabilities

Semantic Search (RAG)

pgvector + OpenAI embeddings

Users can search their knowledge base using natural language. Embeddings enable semantic similarity search beyond keyword matching.

Scheduled Tasks

Cron + FastAPI endpoints

Daily AI briefings, weekly digests, notification delivery. Railway cron jobs trigger HTTP endpoints for background processing.

Rate Limiting

slowapi + Redis (fallback in-memory)

Per-user rate limits on chat (20/hour), list endpoints (100/hour). Prevents abuse while allowing generous usage.

Authentication & Security

JWT + bcrypt + CORS

Secure token-based auth with refresh tokens. Passwords hashed with bcrypt. CORS configured for cross-origin safety.

Email Integration

Resend API + HTML templates

Transactional emails for digests and briefings. Beautiful HTML templates with personalized content.

Real-Time Notifications

Smart surfacing + time-context

Items resurface based on AI-determined time_sensitivity and notification_frequency. Users stay on top of what matters.

Telegram Bot Integration

Webhook + session persistence

Link your Telegram account, capture items on-the-go, and chat with your AI agent directly from Telegram. Full command support: /save, /new, /help.

Why It Matters

This is agentic AI done right

Not a chatbot. Not a simple LLM wrapper. A full-featured agentic system with memory, tools, semantic understanding, and production-grade infrastructure.

Stateful conversations with full context retention
Dynamic tool execution with real-time feedback
Vector embeddings for semantic knowledge retrieval
Scheduled AI-generated briefings and notifications
Type-safe end-to-end with TypeScript + Python
Production deployment on Railway + Vercel + Supabase
agent.py
# Tool-calling agent loop
for event in agent_service.chat(
    session_id=session_id,
    message=user_message,
    db=db,
    user_id=user.id
):
    match event["type"]:
        case "text_delta":
            yield event  # Stream tokens
        case "tool_start":
            yield event  # Tool execution
        case "tool_result":
            if event["mutated"]:
                # Invalidate cache
                ...
        case "done":
            break

Experience it yourself

See what a production-grade agentic AI application feels like. No credit card required.

Try MindStash freeBack to Home