Ask. Answer. Keep working.

Automotive data changes daily: part numbers, availability, prices. A trained model is outdated from day one.

Our live RAG approach always retrieves current data – no training cycles, no stale information.

Additionally: many data providers do not allow training on their data for compliance reasons.

No single provider covers everything: parts catalogs ≠ repair manuals ≠ diagnostic data.

Workshops have preferences: some trust TecAlliance for parts, AutoData for procedures.

The integration is the real work – not the LLM. We also connect workshop systems and workflows.

Agentic LLM: Fast and creative – orchestrates data retrieval, selects sources, plans steps.

Final LLM: Reliable and grounded – generates the answer only from retrieved data.

Result: Cost optimization (expensive model only where it counts) and quality optimization (hallucinations reduced through grounding).

Both LLMs are freely selectable: Gemini, OpenAI, Anthropic, or open-source – no vendor lock-in.

Source grounding: RAG-based – every statement references original data.

Liability: In the automotive sector, wrong information costs money – wrong parts, returns, lost time.

Compliance: Traceability towards customers and manufacturers is often mandatory. No “the AI said it” excuses – instead, trust through transparency.

We've been working for 2 years focused on exactly this problem – understanding workshop workflows, analyzing real mechanic questions, building integrations that work in practice.

Could someone replicate this? In theory, yes. But they would need:

• Months to rebuild our integrations – plus chat interface, multi-modality, deployment infrastructure, and all the domain-specific nuances in our Context Manager
• A team that combines automotive experience with AI engineering (rare)
• Time to build the data flywheel we already have – every query improves our system

Meanwhile, we build more integrations, learn from real usage, and solidify customer relationships. Our lead grows – it doesn't shrink.

Parts catalog integration: 8–15 person-days.

This includes: API integration, mapping to your DMS, testing phase with real data, and fine-tuning answer quality.

Additional data sources (repair manuals, diagnostic data) can be added in parallel or incrementally.

Klarstand works with different part numbering systems – depending on what the connected data source provides. Whether generic article number, OE number, or free-text description: the system automatically identifies available information and matches accordingly.

Where structured numbers are missing, the LLM bridges the gap through intelligent matching of text descriptions – based on existing data, not guesswork.

Currently the focus is on the repair process – from fault diagnosis to repair instructions, technical data, and spare parts. This is a well-defined, high-value use case.

But Klarstand is not limited to repair data. Wherever data from multiple systems comes together, Klarstand can help: vehicle check-in, parts ordering, invoicing. The goal: speed up search and reduce friction between systems.

Data minimization: Only operationally necessary data is collected.

Encryption: In transit and at rest.

No sharing with third parties (exception: LLM processing). User-initiated deletion available at any time. No IP addresses stored for analytics.

EU data residency and on-premise options actively evaluated for production.

Our model is transparently composed of:

• One-time setup fee (based on integration effort)
• Recurring license fee (scaled by users or workshops)
• Maintenance flat rate
• Optional service fee

As a startup, we're flexible and open to creative partnership models. Concrete pricing depends on the desired integration depth.

Domain-specific system instructions restrict the chatbot to the professional context.

For critical facts (torques, intervals, part information), the system must use verified data sources – guessing is explicitly not allowed.

Source references are visible in the UI.

Planned: Automated validation of critical answers against underlying documents – unverifiable answers are discarded.

Yes, our architecture supports it – no vendor lock-in. Open-source and local models deliver usable results but lag behind cloud models in capability.

For maximum quality: Cloud models (Gemini, OpenAI, Anthropic).

For maximum data sovereignty: On-premise operation with local models.

Open-source model quality is steadily improving.

REST API is already available. Bidirectional integration is possible: ingest data into Klarstand and integrate Klarstand functions into partner systems (chat, diagnostic workflows, RMI retrieval).

Supports vehicle databases, maintenance and diagnostic data, parts catalogs, and document retrieval. Additional protocols on request.