Custom abstraction layer supporting multiple AI providers (OpenAI, Anthropic, Gemini, Mistral, Grok, Azure OpenAI). Administrators can configure and swap models dynamically. Unified interface regardless of the underlying provider.

Application deployed across on-prem, AWS SaaS, and Azure (client infrastructure) On-prem and Azure deployments use Azure OpenAI hosted in Sweden (EU/GDPR zone) SaaS users can choose between GDPR-compliant and non-EU models Infrastructure as Code approach: Pulumi (SaaS), Docker Compose (on-prem), ARM templates (Azure)

AI Agent capable of multi-step planning, tool chaining, and iterative reasoning Integration with MCP (Model Context Protocol) servers – both custom-built and off-the-shelf MCP connects to enterprise systems (ERP/CRM databases and applications) Conversational context management across complex interactions

Retrieval-Augmented Generation over millions of corporate documents Multiple chunking approaches: fixed-size, overlap, single-document, semantic chunking Vector search using Pinecone or Azure Cognitive Search as vector stores Integration with Apache Solr for keyword-based retrieval

Full multitenant architecture across both products (Index & Agent) ACL-based security trimming – search results filtered by user permissions Tenant-level configuration of AI models, data sources, and integrations Scalable to thousands of users with 100–1,000 concurrent real-time sessions

~15 microservices (Index) and ~5-6 microservices (Agent), containerized with Docker Custom-built connectors/crawlers for SharePoint, Exchange, ERP, databases, file systems Near real-time synchronization of corporate data sources REST API-based inter-service communication