The technology behind Smartegy.ai
Discover the agent-based architecture that enables natural language business data analysis with over 51 specialized AI agents.
Platform overview
Smartegy.ai is an enterprise-grade, AI-powered business intelligence platform designed for the B2B sector. The system enables business data analysis, visualization, and export through a natural language interface.
Each client receives their own isolated workspace with personalized data sources, permissions, and AI agents. The platform is multilingual and fully GDPR-compliant.
System architecture
Smartegy.ai employs a multi-layered, agent-based (agentic) architecture. The user formulates their request in natural language, which the orchestrator interprets and delegates to the appropriate specialized agents.
Agent-based operation
The "agent-based" (agentic) approach means that the system is not a single monolithic AI, but a collaboration of specialized agents. Each agent is optimized for a specific task domain.
The orchestrator (central AI engine) analyzes the user's request, then automatically selects and invokes the necessary agents. For complex queries, multiple agents may collaborate with each other.
Request reception
The user formulates their question or instruction in natural language through the chat interface.
Orchestrator analysis
The AI engine interprets the request, identifies required operations, and selects the appropriate agents.
Agent execution
The selected agents execute their tasks: SQL queries, report generation, email sending, etc.
Response compilation
The orchestrator summarizes the agents' results and provides a natural language response to the user.
Deterministic execution
Although AI interpretation is based on a language model, agent execution is deterministic: an SQL query always returns precise results, an email always goes to exactly the specified recipient. The language model does not "hallucinate" data — agents work from real sources.
Agent catalog
Smartegy.ai has over 51 specialized agents that can be activated according to each client's needs. All agents are protected by an authorization system.
Book generator agent
Create multi-chapter, book-format PDF documents. Cover page, automatic table of contents, chapters with images and charts. Ideal for annual reports and manuals.
Browser Automation Agent
Automated web interaction: data collection, screenshots, form filling. RPA (Robotic Process Automation) capability.
Business Calendar Agent
Manage holidays, working days, and school breaks. Support business planning and scheduling.
Calculator Agent
Execute precise mathematical calculations. LLMs alone are unreliable at arithmetic — this agent guarantees accurate results for every computation.
Chart & Visualization Agent
Interactive chart creation: bar, line, pie, area, heatmap. Interactive map generation with geocoding.
Cloud Storage Agent
Access remote cloud services (e.g. Dropbox, Google Drive, OneDrive). Download, upload, and sync files directly from the AI workspace.
Competitor Analysis Agent
Comprehensive competitor analysis: industry context, market position, financial data, and strategic comparison.
Contact Manager Agent
Personal contact list with smart duplicate detection and fuzzy search. Store names, emails, phones, companies, tags and notes — all per-user isolated.
Contract Manager Agent
Full contract lifecycle management: creation, search, status changes, expiry alerts, document attachments and access control. Semantic and full-text hybrid search with audit trail.
Custom API Controller Agent
Call various external and internal APIs using natural language instructions. Virtually any local system becomes controllable through AI.
Database Query Agent
Automatic SQL query generation from natural language. Read-only (SELECT) access, schema-aware operation. Supports MariaDB and PostgreSQL databases.
Date Handler Agent
Date calculations, timezone management, and business day awareness. Eliminates LLM date calculation inaccuracies, always delivers reliable results.
Document editor agent
Interactive text editor where you and the AI collaborate on a document in real-time. Writing, editing, commenting, templates, and status management.
Document generator agent
Create business documents with AI assistance: proposals, contracts, letters, minutes. The AI writes it, you approve and download the finished file.
Document parser agent
Extract text from any document: PDF, Word, Excel, PowerPoint, images. Scanned and image-based PDF recognition with OCR (Azure AI + Claude Vision).
Economic Data Agent
Access macroeconomic indicators: MNB exchange rates, KSH statistics, Eurostat and FRED data from a single interface.
Email Agent
Email sending with attachments (PDF, Excel reports). Intelligent recipient resolution by name. HTML-formatted messages.
Excel spreadsheet agent
Create formatted Excel (.xlsx) workbooks from data or database queries. Multiple worksheets, conditional formatting, filters, and charts.
File manager agent
Browse, read, and manage files in the knowledge base, uploads, and personal file folder. Access Word, Excel, PDF, and text file contents.
Flowchart Builder Agent
Create interactive flowcharts, org charts, and mind maps from natural language. Visualize process steps, decision points, and swimlane diagrams — with export support.
Forecast & Prediction Agent
Time-series forecasting for sales, financial or logistics data. Confidence intervals and seasonality detection included.
Gmail Integration Agent
Gmail account integration: read, search, and process emails directly from the AI chat.
Help and guide agent
Built-in user manual and guide. Ask how any feature works and get an instant answer.
Image Generation Agent
AI-based image generation and editing from text instructions. Create marketing materials and visual content.
IMAP Email Agent
Multi-account email client: read, search, send, and manage emails across multiple IMAP/SMTP accounts directly from AI chat.
Interactive Map Agent
Visualize business data on maps: store locator, heatmaps and hexbin layers. Automatic geocoding with Kepler.gl visualization.
Knowledge Search (RAG) Agent
Semantic search across your document library using vector embeddings. Upload PDFs, Word, or Excel files and get answers with source attribution.
Local file manager agent
Access files on your computer directly from the AI chat. Read, write, search, and delete files through the Smartegy Desktop application.
Memory Manager Agent
Persistent AI memory: the system remembers context from previous conversations, preferences, and decisions. Per-user isolated, long-term knowledge base.
My Files Agent
Browse and download generated files in an interactive panel. Excel, PDF, charts, PowerPoint and Word files in one place — convenient overview and reuse.
Orchestrator agent
Coordinate complex, multi-step tasks with persistent state management. Prevents context window overflow and enables resuming interrupted tasks.
PDF report agent
Quick, dashboard-style PDF reports with charts, KPI indicators, and tables. Single or multi-page summaries with multi-column layout support.
Personal Calendar Agent
Create, edit and search calendar events. Set reminders via email, SMS or in-app notifications. Recurring events and interactive month/week view.
Personal Dashboard Agent
Custom dashboard with drag-and-drop widgets: KPIs, charts, database queries, iframes, and live data — your business at a glance.
PowerPoint presentation agent
Professional PowerPoint (.pptx) presentations with editable charts, KPIs, and multiple slide types. 5 built-in themes, corporate branding support.
Project Manager Agent
Full-featured project management: tasks, sprints, Gantt charts, time tracking, knowledge base, and team collaboration — all managed through AI chat.
RFM Customer Segmentation Agent
Automatic customer segmentation using Recency-Frequency-Monetary analysis. Ideal for campaign targeting and customer retention.
Scheduling Agent
Schedule recurring, automated tasks: daily reports, weekly summaries, regular analyses running in the background.
Semantic Memory Agent
Deep AI memory with semantic search: reminders, appointments, conversation context, AI self-reflection, and long-term insights — all per-user isolated.
Smart Prediction Agent
Context-aware follow-up suggestions: the AI anticipates your next question based on conversation history and provides clickable predictions.
Smart Recipient Resolver Agent
Automatically resolves names to email recipients using AI with context-aware matching. Requests confirmation when multiple matches are found.
SMS Agent
Send SMS messages directly from chat. Automate business notifications and alerts.
Support Ticket Agent
Ticketing has never been this easy. For any issue, the AI automatically opens a ticket with the support team — just describe the problem.
Task Manager (Kanban) Agent
Personal task management with interactive Kanban board. Create, organize, prioritize and track your tasks — with drag-and-drop columns and fuzzy search.
Translator agent
Professional document and text translation in 30+ languages. Translate PDF, Word, PowerPoint, and Excel files while preserving the original formatting.
Video Generation Agent
Create short videos with AI assistance. Generate marketing content and presentation materials.
Weather Agent
16-day forecast and 80 years of historical weather data. Useful for agricultural, logistics, and retail planning.
Web Search Agent
Real-time internet search and web content retrieval. Access to current information, news, and market data.
WhatsApp Agent
Send WhatsApp messages directly from AI chat. Smart name-to-phone resolution, template messages, and human-in-the-loop approval before sending.
Word document agent
Generate native Word (.docx) files with charts, KPI indicators, tables, and templates. The document is editable in Word or Google Docs.
Workflow Orchestrator Agent
Build and automate any business process. Human-in-the-loop capability: request human approval at critical points before proceeding.
Custom Agents — Unlimited Possibilities
The agents above are just the starting point. Any specialized agent can be developed and integrated based on your unique requirements.
Custom agent examples
Camera Agent
Image recognition and visual analysis. The AI sees: product identification, quality control, inventory management through cameras.
IoT Agent
Communication with various sensors and IoT devices. Temperature, humidity, motion detection — real-time data collection and response.
and any custom development tailored to your needs...
Complete ERP Functionality — AI-Powered
With Smartegy.ai's agent-based architecture, full enterprise resource planning becomes reality. Even the most complex custom tasks can be personalized and automated.
Human-in-the-Loop
Request human approval at critical decisions. The AI suggests, the human decides — full control at every important step.
Audit & Control
Every operation logged, every decision traceable. Complete transparency and compliance guaranteed.
Fully Customizable
No two businesses are alike. Agents can be freely combined, extended, and tailored to your unique workflows.
This is not the future — it is already the present. AI-powered ERP that adapts to you.
You always decide — AI only helps you decide and execute
The AI never acts alone on critical operations. Every sensitive action requires explicit human approval — this is not an AI decision, it is a deterministic, rule-based safeguard configurable per company and per user.
AI agents prepare and propose — the human reviews and approves before execution
AI errors cannot happen
The Human-in-the-Loop system is not a recommendation — it is a hard blocker. When enabled for an operation, the AI physically cannot execute it without human approval. This is not the AI deciding to ask — it is a deterministic rule that makes unauthorized execution impossible.
The approval workflow is enforced at the system level, not by the language model. Even if the AI "wanted to" skip it, the architecture prevents it. The LLM proposes, the deterministic system enforces, and the human decides.
Critical operation protection
Emails, SMS messages, database modifications, file operations, financial transactions — any action with real-world consequences can be gated behind mandatory human approval.
Per-company & per-user configuration
Each company defines its own approval rules. Within a company, different users can have different approval requirements. A manager may auto-approve what a junior employee needs explicit confirmation for.
Deterministic, not AI-decided
The approval requirement is a system-level rule, not an AI judgment call. The language model cannot override, bypass, or "forget" to ask. If the rule says "require approval" — approval is required, period.
Fully customizable approval flows
Define exactly which operations need approval, for which users, under what conditions. Set thresholds, combine with audit logging, and create multi-step approval chains for the most sensitive workflows.
Dedicated EU GPU server & self-hosted AI models
Beyond the large language model (LLM), Smartegy.ai runs specialized AI models on its own EU-based rented GPU server. These models never leave our infrastructure — they operate exclusively within the EU under our own supervision.
Forecasting, entity recognition, and text processing models run on dedicated GPUs, ensuring fast response times and EU-only data processing.
Exclusively EU-based processing
All self-hosted AI models run on an EU-based rented GPU server. Data never leaves the EU territory.
Business forecasting models
Six different forecasting models are available for time-series based business planning. Automatic detection of sales trends, seasonality, holiday effects, and anomalies.
NER — Named Entity Recognition
Automatic recognition and anonymization of person names, company names, location names, and other sensitive entities in business data. The foundation of the anonymization feature.
Chronos — Neural forecasting
Amazon Chronos neural foundation model with LoRA fine-tuning. Company-specific adapted models: more accurate than general forecasting, trained on each company's own data patterns.
Number-to-text conversion
Number-to-text conversion service supporting all system languages. Currencies, percentages, units, and tables transformed into speech-optimized text.
Custom large language model (LLM) deployment
Upon request, it is possible to deploy and operate a custom large language model on the rented EU GPU server. This keeps the entire AI operation within your own infrastructure, without involving third parties.
This option is available exclusively under a permanent contract and individual agreement, and is not available during the pilot period.
Voice control & speech synthesis
Smartegy.ai offers full voice-based interaction: speech recognition for dictating questions and natural-sounding text-to-speech for AI responses.
Google Speech Recognition
Real-time speech recognition directly in the browser. Users click the microphone button to dictate their questions, which the system converts to text and sends to the AI. Supports all system languages.
ElevenLabs text-to-speech
Natural-sounding AI response readout powered by ElevenLabs technology. Streaming-based speech synthesis: the AI response arrives sentence by sentence and becomes audible immediately with minimal latency.
Data protection notice for ElevenLabs service
When using ElevenLabs text-to-speech, the response text is processed on ElevenLabs servers. Data is deleted after use. Users can disable voice readout at any time. Speech recognition (Google Speech) runs in the browser — no audio data is stored on our servers.
Security & data protection
The foundation of Smartegy.ai's architecture is a multi-layered security model. Every client works in a fully isolated environment, exclusively on EU servers.
Full tenant isolation
Each client runs in their own containerized environment. Data and workspaces are physically separated from each other.
End-to-end encryption
All communication is protected with TLS 1.3 encryption. Data is encrypted at rest as well.
No AI model training
Client data is NEVER used for AI model training. Data is used exclusively to answer the user's own queries.
GDPR compliance
Fully compliant with the EU General Data Protection Regulation. All data processing occurs exclusively within the EU.
Audit logging
Every user action is logged: who, when, what was asked, and which agent was used. 90-day retention.
Permission system
Three-tier permission management: company level, category level, and user level. Real-time updates.
Two-Factor Authentication (2FA)
Smartegy.ai supports two-factor authentication (2FA), which significantly enhances account security. After enabling 2FA, logging in requires both a password and a time-based one-time password (TOTP).
Activation
The user activates 2FA in their settings. The system generates a QR code.
Authenticator App
The user scans the QR code with a TOTP-compatible authenticator app (e.g., Google Authenticator, Microsoft Authenticator, Authy).
Verification
The user enters the 6-digit code generated by the app to confirm activation.
Login with 2FA
From then on, each login requires entering the current 6-digit code from the authenticator app after the password.
Technology stack
Smartegy.ai runs exclusively on EU infrastructure, meeting the highest data protection and performance requirements.
Server infrastructure
Dedicated physical server at Scaleway's French data center. France was chosen because Scaleway provides dedicated hardware with full physical control — not shared cloud VMs, but a self-managed server. Scaleway is a major European provider ensuring full EU data sovereignty.
AI processing
AWS Bedrock service in the Frankfurt (eu-central-1) region. EU inference profile, no data retention, no model training.
Leading global financial institutions run on AWS Bedrock + Claude:
Citi · RBC Capital Markets · Visa · AIG · Moody's · Commonwealth Bank of Australia · NBIM · Bridgewater · Coinbase · LSEG
View references on Anthropic's websiteLanguage model
Anthropic Claude models — the most advanced language models for business analysis, code generation, and natural language interaction. Any LLM can be used as orchestrator or sub-orchestrator upon client request (e.g. GPT, Gemini, Mistral, Llama, or even self-hosted models).
Token Optimization & Cost-Efficient Operation
Smartegy.ai uses proprietary, specialized token optimization technology to operate LLMs cost-effectively and efficiently — without compromising response quality.
Intelligent prompt caching
The system automatically caches static system messages (system prompt, agent definitions), so the same context is not reprocessed with every request. This results in up to 90% cost savings on input tokens.
Session compaction
During longer conversations, earlier messages undergo intelligent compaction. The system preserves essential information while drastically reducing context size.
Intelligent model routing
The system automatically selects the appropriate model for each task: smaller, faster and cheaper models for simple subtasks, and the most powerful model for complex analyses.
Flexible LLM Selection
Any large language model can be configured as orchestrator or sub-orchestrator upon client request. The provider-agnostic architecture supports Anthropic Claude, OpenAI GPT, Google Gemini, Mistral, Meta Llama, or even self-hosted models.
Model switching is done at configuration level, without application changes.
Data Backup & Recovery
Every customer instance, audit log, and stored data is automatically backed up daily. Backups are stored geo-redundantly at Hetzner's data center in Germany.
Daily Backup
Automatic nightly backup. The previous day's state can be restored.
Hetzner, Germany (EU)
Geo-redundant storage at Hetzner's German data center, within the EU.
Comprehensive
Every application instance, database, audit log, and user data is backed up.
Recovery
The previous day's state can be restored. Smartegy.ai assumes no liability for data loss.
Smartegy.ai handles backups with the utmost care; however, we assume no liability for data loss occurring between backup cycles.