AI that knows your data.And never leaves your network.
Most AI tools do not fail on the technology but on the question of where the data goes. fabAI solves that differently: the language model is embedded in fabular and runs on your hardware – with access to every order, every batch, every supplier.
- AI agents from day one
- Workflows built in dialogue
- Fully auditable
- 100%
- runnable locally
- 21
- ready-made agents
- −68%
- fewer routine tasks
- 0.3 s
- average response time
No cloud lock-in, no per-request API cost
Usable from day one
Measured at existing customers
In ERP context, on local hardware
Three reasons AI gets stuck in mid-sized companies
It is almost always the same three – and all of them are about control, not about model quality.
Sensitive data is supposed to go to the cloud
Processing production, customer and supplier data in an external AI service? For many companies that is not an option, legally or strategically – and the EU AI Act does not make the justification any easier.
fabAI runs entirely locally. Not a single byte leaves your network.
The AI does not know the business
A generic model knows nothing about your recipes, your batch logic or your supplier terms. The answers stay generic and therefore worthless in daily work.
fabAI works directly on your ERP data – with permission checks.
Nobody can constrain the AI
External services are black boxes: no control over what the model may do, who sees the queries and what gets logged. For an audit that is a problem.
Your IT defines rights, limits and logging – traceably.
Agents that finish the job
21 ready-made agents are usable from day one – from chat to actually closing out a case, not just answering it.
Agents instead of chat
fabAI carries out actions: it creates records, builds entire report pages and thinks along on master data screens.
Process documents automatically
Invoices, delivery notes and forms are read, classified and linked to the matching ERP record.
Reorder proposals
Consumption patterns and stock levels are analysed, timing and quantity proposed proactively.
Anomaly detection
Deviations in production, batches or processes stand out automatically – before they turn into an incident.
Meeting-minutes agent
Transcript in, minutes out: the summary is filed against the partner in the CRM and action points are derived.
Reports & forecasts
Reports, sales forecasts and key figures are produced automatically and land directly in your dashboards.
All 21 ready-made agents at a glance – organised by business area.
Sales & eCommerce
Sales Funnel Agent
Guides prospects through the sales funnel and flags where a funnel stalls.
X-selling Agent
Suggests matching add-on and alternative items for the current basket.
Product Knowledge Agent
Answers product questions from the real article data, not a separate wiki.
Sales Agent
Supports sales with enquiries, quotes and following up open opportunities.
Support Agent
Answers customer queries based on order and history data.
Order Calculation Agent
Calculates quotes and order lines, including scales and terms.
Sales Forecasting Agent
Derives sales forecasts from actual sales and consumption data.
Order Processing Agent
Checks and processes incoming orders, from entry through to release.
Logistics, delivery & purchasing
Supply Chain Agent
Monitors supply chains and flags supply risks before a shortage occurs.
Internal Logistics Agent
Coordinates internal goods movements between warehouse and production.
External Transport Agent
Plans and tracks delivery to customers via the connected carriers.
Purchasing Agent
Creates reorder proposals from demand, stock and lead times.
Incoming Invoice Processing Agent
Reads incoming invoices, checks them against the order and goods receipt, and assigns them.
Production, maintenance & quality
Production Resource Planning Agent
Plans capacities, orders and resources in production.
Infrastructure KnowHow Agent
Keeps maintenance knowledge and asset history at hand when a fault occurs.
Quality Customization Agent
Adapts inspection plans and quality parameters to customer-specific requirements.
Finance, communication & knowledge
Invoicing Agent
Creates invoices from the delivered or rendered scope.
Booking Agent
Posts recurring and rule-based transactions in financial accounting.
Periode Control Agent
Checks period closings for completeness and anomalies.
Postman Agent
Sends notifications and messages automatically to the right place.
Knowledge Agent
Answers questions about company-wide knowledge from documents and filing (ECM).
In preparation: an open fabAI agent marketplace where customers will be able to share their own agents with each other – alongside the 21 ready-made agents included as standard.
The same week, distributed differently
The figure of −68% routine tasks stays abstract as long as you cannot see what it does to a working day. The bars are an example calculation based on that reduction – how the time is distributed in an individual case depends on the process and the starting point.
Example: without AI support
Assume that two thirds of the time goes on entering, searching and chasing up.
- Routine: entering, assigning, chasing up
- 68 %
- Specialist work and decisions
- 32 %
The same week with fabAI
At −68% routine tasks, around a fifth of that remains.
- Routine: entering, assigning, chasing up
- 22 %
- Specialist work and decisions
- 78 %
Build new workflows in dialogue, don't commission them
Your own AI assistants for recurring tasks and your own screens emerge through low-code – described instead of programmed. This is Beyond Vibecoding: newly built, on an audited foundation. Outward-facing, fabAI stays connectable: as an MCP server, fabular is reachable for other providers' AI agents; as an MCP client, fabAI itself addresses the servers of your other systems.
- Skill builder: your own assistants for recurring tasks, without programming
- Custom pages: your own screens and reports on request
- MCP server: fabular is reachable for outside AI agents
- MCP client: fabAI carries out steps in your other systems itself
Skill builder & custom pages
Your own department builds assistants and screens itself – without a development ticket.
MCP in both directions
Server and client at once: fabAI connects to your wider system landscape, not just to itself.
Your data. Your servers. Your AI.
fabAI is the only ERP-integrated AI solution in the DACH region that can run entirely on local hardware – with a permanently embedded language model and no cloud connection. If you prefer cloud models, you can run hybrid, with a clear separation between locally and externally processed data.
- All processing runs on your servers – including the model itself
- No separate setup, no external model provider, no token billing
- Rights, action limits and logging stay with your IT
- Hybrid operation is possible where cloud models are wanted
Embedded language model
A fixed part of fabular, running on your hardware. No outbound connection needed.
Access to real ERP data
Batches, orders, stock, finance – through the same permissions that apply to people.
Auditable from the start
Every query and every action an agent takes is logged and attributed to a user.
Security in detail
Granular permissions
Who may use which AI function is defined by role and at department level – fully auditable.
Complete logging
Every query and every action is attributed to a user and traceable – including for an audit.
ISO 27001 & GDPR
Operated in an ISO 27001-certified data centre or fully on-premises; processing is GDPR-compliant within the EU.
What management and IT each get out of it
fabAI has to pass two tests: it must measurably remove work, and it must stay controllable.
Management
Decisions on current figures
- Real-time insight without manually compiled reports
- Reorder proposals and forecasts on actual consumption data
- Deviations are reported before they escalate
- Summaries and reports on request instead of on request to someone
- Less time in operational detail, more in steering
IT leadership
Control that holds up in an audit
- Local model – data sovereignty stays in the building
- Configure, extend and switch off skills yourself
- Access and action rights role-based and granular
- Every AI action logged and traceable
- No dependency on a third party's price list
fabAI and external AI tools
ChatGPT, Copilot and similar services are capable. They just do not know your data – and do not run in your infrastructure.
| Comparison | fabAIERP-integrated AI | External AI toolsChatGPT, Copilot & co. |
|---|---|---|
| Processing on your own hardware | Fully local possible | Cloud processing |
| Access to ERP data and processes | Full, via existing permissions | No access |
| Extensible AI skills | Skill import and skill builder | Limited, via API |
| Data sovereignty and GDPR | Guaranteed on-premises | Not assured |
| IT control over rules and access | Fully configurable | No control |
| Running cost per request | Model included | Billed per use |
| Complete logging | Auditable | Black box |
| Industry-specific process logic | 20 industry solutions as context | Generic |
The comparison refers to standard operation of each service. Enterprise agreements with external providers can soften individual points but do not remove cloud processing.
What you should know about fabAI
What makes fabAI a better alternative to ChatGPT or Microsoft Copilot for a business?
fabAI is embedded in the ERP and can run entirely on your own hardware – ChatGPT and Copilot process in the cloud and do not know your order, batch and stock data. Because fabAI uses the same permissions as the ERP and logs every action, it holds up to an audit, while external services stay a black box in that regard. There is no running per-request cost, because the model is included as standard instead of billed by usage.
What hardware do we need for local operation?
For the embedded model a server with a current GPU is usually enough – we size it together, based on your number of users and the planned use cases. If you do not want to buy your own hardware, you can run fabAI hybrid or start with a cloud model and move to local operation later.
How does fabAI make sure nobody sees data they are not allowed to see?
fabAI uses the same permission checks as the interface. A query returns only data the signed-in user would also have access to in the ERP. On top of that, your IT defines which actions an agent may perform at all – read, create, change – and every action is logged.
What is the difference between a skill and an agent?
A skill is a bounded capability, for example reading an incoming invoice and assigning it. An agent combines skills into a sequence and works through it on its own, for example the meeting-minutes agent: understand the transcript, summarise it, file it in the CRM, derive action points. You build skills yourself in the skill builder, without programming.
And if we do not use fabular at all?
For that case there is the enterprise AI solution: AI-supported analysis and your own assistants on your data, independent of your existing ERP system. That is often the first step before a platform decision is discussed.
How does fabAI deal with the EU AI Act?
Local operation simplifies the classification considerably, because no data is transmitted to an external provider and logging stays entirely with you. Which obligations apply to your specific use cases is something we clarify in the AI consulting engagement – including documentation and governance.
See fabAI on real data
We do not demonstrate the agents on a sample dataset but on a case from your business – and we say openly where AI contributes nothing yet.