Concept

dai.guru aims to replace our existing ERP system step by step with an AI-supported system. AI will support individual users and also organise the ongoing development of the software.

Customers’ day-to-day work provides the essential input. They can send requirements, feedback, specific tasks and simple questions to GURU by email or Telegram. With central access to system knowledge, documentation and source code, GURU can answer many questions directly. New requirements lead to traceable change proposals that are developed, documented and checked in a safe test environment.

From a requirement to a new feature

1. Questions and customer requirements as the starting point

Customers continue working with their ERP system as usual. They can communicate with GURU by email or Telegram. Not every message needs to become a development task: customers and administrators can also ask GURU about operation, workflows, data structures or system behaviour. Access permissions still need to be clarified.

GURU uses its knowledge of the installation, together with the available documentation and source code, to provide informed answers. If it identifies a bug, a missing feature or an opportunity for improvement, this can become a documented change proposal for the development and testing process.

This way, the system evolves alongside the actual requirements of each installation.

2. The central GURU VM

The GURU VM is the central environment for development and system knowledge. It provides the necessary information, source code, tools and compilers. GURU analyses new requirements, develops proposed solutions and creates the corresponding tests and documentation.

All changes are versioned in GitHub repositories. Source code, decisions and development history therefore remain traceable at all times.

3. A test VM for each installation

Each customer installation has its own virtual test machine. It reproduces the working environment using test data, without production customer data. An approved PC at the customer’s premises can also serve this purpose.

Each test VM or PC runs its own AI instance, Claude or Codex. This AI test agent is responsible exclusively for that customer’s environment. It knows the configuration, special features and test procedures, and coordinates the installation and testing of changes developed by GURU.

The AI test agent runs the planned tests with test data, identifies discrepancies, documents the results and reports them to the central GURU VM. If a test fails, the proposed solution returns to GURU with the collected information for further work. The production installation remains unchanged.

The customer-specific AI does not decide whether to release a change. It prepares the technical checks and their documentation; approval for the production system remains with the responsible system administrator.

4. The customer decides

Only after the checks have passed is the customer’s responsible system administrator notified. The administrator receives the results and decides whether and when the new feature will be installed in the production environment.

Guiding principles

• Driven by practical needs: Real customer requirements guide development.
• Directly accessible: Questions and tasks can be sent to GURU by email or Telegram.
• Based on system knowledge: GURU answers questions using documentation, installation knowledge and source code.
• Centrally coordinated: GURU brings together knowledge, tools, development and documentation.
• Tested for each customer: Every test VM has its own AI test agent dedicated to that customer’s environment.
• Traceable: Every change is versioned and documented in GitHub.
• Safely tested: New features first pass through the appropriate customer test environment.
• Released under human control: No change reaches production without the system administrator’s decision.

The result is an ERP system that can learn and evolve, continuously adapting to new requirements while retaining control, testability and traceability.