DtoMES combines 40 years of ERP/MES experience with a modern architecture of BPL packages, interfaces and thread-safe pools. Multiple databases orchestrated by synchronization, with PostgreSQL as the platform & AI hub, Ollama for local embeddings, and modules that talk to AI through MCP.
An informative side to understand what the product does and who it is for; a technical side for those who want to see how it is built. Same platform, two levels of reading.
The platform, the verticals (manufacturing and associations), why DtoMES, our story and contacts. Value before the details.
Platform · Verticals · Why →BPL architecture, PostgreSQL as the data hub, local AI with Ollama and MCP, modules (Production, Planning, Laboratory, Documents), MemoryLens, security and audit.
Architecture · AI & MCP · Modules →Each module is an independent BPL: it compiles, tests and ships on its own. Updating a module means replacing one .bpl file. Instant rollback.
The platform is multi-database by design: Firebird for the MES, SQL Server/COBOL for the ERP, PostgreSQL as the platform & AI hub. Hash-based synchronization keeps them aligned, with no triggers on source tables.
Not a chatbot in a corner. Embeddings run locally with Ollama; modules expose and consume AI tools via MCP. AI becomes infrastructure.
Centralized RBAC, declarative enforcement on forms and actions, access audit on PostgreSQL, separation of duties. Foundations ready for future reviews.
Every piece of data leaves a trail. Audit log and append-only hash chain: if a result is wrong, you can trace the cause event by event.
Common management, local accounting: adding a country is configuration, not rewriting. A neutral accounting contract with a per-country profile.
The platform's data layer converges on PostgreSQL (with
pgvector for embeddings),
while synchronization keeps the operational databases aligned. Semantic intelligence runs locally with
Ollama; integration with AI models goes through MCP, the protocol that exposes modules as tools.
The hub for platform services: identity, permissions, telemetry and documentation. With pgvector it also holds embeddings: semantic search next to the data, without a second engine.
Embedding generation runs locally, without sending sensitive data outside. Documents, notes and knowledge base become vectors for semantic search and contextual support.
Model Context Protocol: modules expose tools (documents, modules, scaffolding, archive) that AI models can use safely and traceably. It is the bridge between the platform and AI.
Bill of materials, barcode, order fulfilment, pallets, hours. MVP with 30 catalogued queries and MemoryLens built in.
Gantt, scheduler, resources, material commitment. Multi-frame host with two operating modes.
MVP controller, import, management, Type A/B detail plugins. FastReport printing.
Plugin architecture for detail types, scanner import, printing and labels.
The historic core: production, planning, laboratory, documents, synchronization with the legacy ERP. The domain where the platform is already in production.
Go to verticals →Members, fees, book sales, events, accounting and compliance. Common management, local accounting — the same multi-country model.
Associations vertical →From a revolutionary idea in 1985 to today's AI-powered future.
A “virtual machine” written in C for SCO Unix that read text configuration files and generated a complete ERP system. Configuration instead of code — and ten years of continuous operation.
Windows and Delphi changed the industry's paradigm. We adapted, but the data-driven DNA stayed in every system we built.
ERP, MES, PLM and Field Service systems for manufacturing and services. Systems still running today and still generating value.
DtoMES / DTO AI Framework: the 1985 vision realized with 2025 technology. BPL architecture, PostgreSQL as the hub, local embeddings with Ollama and AI integrated via MCP.
A short book (~23 pages) on the philosophy of an AI-integrated business system: secure, controlled and documented by design. Interfaces, MVP pattern, MCP, integration and multi-country — explained for the value, not the code.
40 years of experience. Modern architecture. Data on PostgreSQL. AI built in via MCP.