Skip to content

AI layer · agents in the workflows

Artificial intelligence that works inside your tools.

AI that moves customers to purchase: it prioritises the right customers, activates stock and guides the team — while your data stays under your control. Nebula AI moves work from static data views to AI-assisted prioritisation: background assistants turn CRM, inventory and service data into proactive sales actions, and decision support is embedded directly into the workflow tools. Agents can be guided — with human oversight at every step — or find the path to a solution on their own.

4

prototypes in development

Nebula AI — sketch: tool automation, AI workflows, guided and autonomous agents

Who it is for

Importers, dealers and finance institutions on the VEGA and GAIA platforms that want more sales, after-sales and better team utilisation from existing data — without switching tools.

Integrations

VEGA · GAIA · configurators and price lists · call centre · DMS and appointment planner (MPO) · external and internal LLMs · RAG and knowledge graphs · human oversight

Core capabilities

Golden Moment Proactive Sales (VEGA AI) — the right customer at the right time, with an AI-ranked opportunity queue (Hot/Warm/Cold) and a recommended next step.

Stock Management (VEGA AI) — natural-language inventory search, best-fit matching, aging-stock and pricing signals.

Virtual Assistant + interactive learning (VEGA AI) — one agent for sourced answers, guided help, quizzes for easy learning, news and built-in user-support requests — without leaving the CRM.

New Vehicle Sales Targeting (GAIA AI) — ranking customers by purchase probability with recommended timing and offer.

Spare Parts Inventory Activation (GAIA AI) — dead parts stock becomes a targeted after-sales campaign. Planned.

New Vehicle Fleet Sales Agent (GAIA AI) — an agent supporting new-vehicle sales to fleets and key accounts. Planned.

AI Service Booking Assistant — service appointment booking with smart customer identification and slot reservation. Planned.

External or internal LLMs — you choose where the models run: a public provider or your own infrastructure, driven by data sovereignty and security.

Regulated flows and autonomous agents — from strictly governed processes to agents that find their own path to solving a given task, always with human oversight.

Knowledge graphs — advanced knowledge bases built on knowledge graphs for verifiable, grounded answers from data, documents and processes.

Status · August 2026

Four of the seven use cases already run as prototypes in development.

Four of the seven use cases in the Nebula AI portfolio already run as working prototypes in active development, exercised on real automotive data — an assistant that answers from the official manuals with citations and screenshots, natural-language search over new-vehicle stock, and two multi-step pipelines that turn dealer data into ranked, GDPR-gated worklists. Three use cases are planned.

VEGA Asistent — one in-app agent: grounded answers from the manuals with citations and images, quizzes for easy learning, guided troubleshooting, dealer-hub news and built-in requests.

Natural-language stock search — a sentence as a query over the entire current new-vehicle stock.

Golden Moment proactive sales (GM) — customer-first: internal signals → customer ranking (Hot / Warm / Cold) → matching against own stock → offer brief.

New-vehicle sales targeting (NV) — vehicle-first: vehicle ranking (aging, margin, demand) → selecting the right customers → channel by consent.

Planned: an agent for new-vehicle sales to fleets and key accounts, an AI service-booking assistant, and spare-parts inventory activation (AFS).

Prototype in development · VEGA AI

One assistant inside the CRM: answers, learning and support.

VEGA Asistent is a single agent embedded in the corner of the VEGA screen, accompanying the user through everyday work. A question asked in everyday words — »zakaj mi ne pusti shranit ponudbe?« (»why won't it let me save the offer?«) — is first rewritten into the manual's register and answered exclusively from the official VEGA and mVEGA manuals: every sentence cites the page it came from, and the answer is completed by the original screenshots of the cited pages. When the manuals don't cover something, it says so — and takes the user forward another way.

VEGA Asistent — grounded answer flow: a question in everyday words → rewrite into the manual's register → hybrid retrieval → an answer with citations and screenshots from cited pages

Advanced grounded answers with images — from the full library of official manuals in Slovenian and Croatian.

Guided troubleshooting: steps from the manuals that walk the user out of a concrete blocker.

Interactive learning: short quizzes and tips from the manuals, tailored to the user's role — the agent infers the role in the background from how the application is used; nobody fills in a profile.

Staying informed: dealer-hub news, ranked by relevance to the user's work.

Built-in requests: when an answer is not enough, the agent prepares a request for the Nebula team straight from the conversation.

Embedded with a single script tag; the widget is fully isolated and does not interfere with the application.

Prototype in development · VEGA AI

»A family SUV under €30,000 with an automatic« is a valid query.

Salespeople search the entire current stock of Renault, Dacia and Alpine new vehicles by describing what the customer needs, in Slovenian, Croatian or English. An LLM compiles the sentence into a precise, auditable filter (body type, powertrain, price, equipment) over the real factory equipment of every version, taken directly from the vehicles system. A second, semantic channel catches what no equipment code can express — »avto za kampiranje« (»a car for camping«) finds the right trims through roof rails and washable interiors. The two channels blend, and a concierge model orders the final shortlist and explains each pick in the user's language.

Natural-language search: one sentence → compiler → exact equipment channel and semantic channel → blend → concierge with an ordered, explained shortlist

Vehicle equipment is never estimated: the filter reads the actual factory equipment lists of each version — for practically the entire stock.

The query compiler picks only from a closed equipment vocabulary — it never guesses.

When a filter matches nothing, the tool clearly says so — it cannot invent stock.

Prototypes in development · GAIA AI and VEGA AI

Pipelines for stock vehicles and people.

Two back-office use cases run as multi-step pipelines over the dealer's own data — every step predefined, no external feeds, no black-box scoring. Each pipeline is packaged as an MCP server, so any approved AI agent can drive it, and each ends at a human: the manager confirms before anything is sent.

Pipeline with a human gate: internal data → multi-step flow → GDPR gate → ranked campaign → a person confirms → outcome log

Golden Moment (GM): finds the customers ready for their next vehicle — Hot / Warm / Cold on internal signals — and matches each hot customer against the dealer's own stock. When nothing fits, the salesperson gets a configuration brief for a new-build offer instead: call this customer, offer this car.

New-vehicle push (NV): vehicle-first — ranks the new cars that most need to move (aging, margin, demand), then finds the right customers for each; a hard ownership-tenure rule and routing to call centre, e-mail or SMS by score and consent.

Shared guardrails: a GDPR consent gate in the data layer, a shared suppression store (no customer is contacted twice across pipelines), every threshold in a business-owned playbook file rather than in code, and outcome logging that feeds the next run.

Planned: a third pipeline — spare-parts inventory activation (AFS), from stock aging and car-parc fitment to routing into campaign, transfer or write-off.

Planned

Three use cases in the plan: fleets, service and parts.

The portfolio continues with three use cases at the planning stage — under the same rules as the prototypes: grounded answers, GDPR guardrails and a person who decides.

An agent for new-vehicle sales to fleets (key accounts) — from signals across the network to an offer brief for the business customer; built on the Key Accounts view in GAIA.

An AI service-booking assistant — the customer identifies themselves through several parameters (for example part of the chassis number plus the model, the registration number or an after-sales invoice number), and where the assistant has access to a scheduling planner — through the DMS or the MPO tool by our partner CPL, which we are jointly rolling out at Summit Motors Ljubljana — it also books the appointment.

Spare-parts inventory activation (AFS, GAIA AI) — a multi-step pipeline: stock aging → fitment against the car parc → scoring → routing to campaign, transfer or write-off.

Reference

Prototypes in development for GA

7 use cases in portfolio · 4 prototypes in development · 3 planned · prototype trials August 2026

Four use cases in development already run as working prototypes, exercised on real data: VEGA Asistent with grounded answers from the official manuals, natural-language search over new-vehicle stock, and the pipelines for Golden Moment sales and new-vehicle push. The starting point was an earlier prototype of AI in car configurators and price-list management tools (GA Ford); planned next are an agent for new-vehicle sales to fleets, an AI service-booking assistant and spare-parts inventory activation. None of this is a production deployment yet.

FAQ

Frequently asked questions.

What is Nebula AI?
Nebula AI is the AI layer developed by Nebula d.o.o., embedding proactive sales prioritisation, stock activation and virtual assistants directly into the existing tools (VEGA / GAIA) since 2026 — with guided and autonomous agents under human oversight and data sovereignty. As of August 2026, four of the seven use cases in the portfolio run as working prototypes in active development, exercised on real automotive data, and three are planned — none is in production yet.
What AI capabilities does Nebula provide?
An AI layer embedded into the existing tools (VEGA / GAIA): proactive sales prioritisation, stock activation, virtual assistants, external or internal LLMs with data sovereignty, guided and autonomous agents with human oversight, and knowledge graphs — 7 use cases in the portfolio, of which 4 run as prototypes in development (August 2026): a grounded manual assistant, natural-language stock search and two multi-step pipelines for proactive sales and new-vehicle targeting; 3 are planned: a fleet sales agent for key accounts, an AI service-booking assistant and spare-parts inventory activation.
Does the Nebula AI assistant ever answer from general knowledge?
No. It answers only from the indexed official manuals and says when they don't cover a question. Every sentence of the answer carries a citation of the manual page it came from, and the screenshots shown are taken from the cited passages only.
Can the Nebula AI pipelines send campaigns automatically?
No — by design. The pipelines prepare ranked lists with reasons, behind a GDPR consent gate and a shared do-not-contact store; a person confirms every export before anything reaches a customer.
What data leaves the company when using Nebula AI?
Queries go to the selected LLM provider; the document corpus, vehicle stock and customer data stay in the dealer's systems. Where data sovereignty requires it, internal LLM deployment is supported behind the same OpenAI-compatible gateway.