Frequently Asked Questions

No. LOKI is for every company with documents, emails, and a database: law firms, retailers, and service providers as much as machine builders. The same goes for HEIMDALL, which guards the path to cloud AI, and for SAGA, which takes down meetings. MINERVA comes in when you manufacture and want AI all the way to the machine.

One is enough. IGNAITE is the platform, and each product runs on its own and can be booked individually: LOKI, the AI knowledge assistant, for every company. MINERVA, the machine AI, for manufacturing, on hardware we build ourselves. HEIMDALL, the AI firewall, is built into LOKI and can also be booked on its own. SAGA, the AI minute-taker, builds on LOKI’s local technology. Combined, LOKI, MINERVA and HEIMDALL share one memory, one role model and one audit log.

No. Every IGNAITE product runs on its own and can be booked individually: LOKI, MINERVA, HEIMDALL or SAGA. Whatever you add fits in. LOKI, MINERVA and HEIMDALL then share one machine, one login, one role model and one audit log.

Because independence starts with the stack. We're not a wrapper around third-party services: owning the stack from chip to answer brings AI right to the machine and makes you independent of vendors.

Custom GPTs are chat interfaces with a system prompt and uploaded documents. On cloud servers. LOKI is an assistant that gets to work (technically, a tool-calling agent): it queries your databases live, generates ready-made Word and PowerPoint documents, and logs every step. On your hardware. If you use Custom GPTs and are happy with them, you have a different use case. Automating actions, and AI that grows with you, needs a different architecture.

Yes, via the REST API. LOKI already has a complete API for documents, collections, databases, and chats. Anyone who wants to build their own frontends, workflows, or integrations can do so. An official skill SDK for tools and workflows is in planning.

HEIMDALL is the firewall for AI, built into LOKI and also bookable on its own. It checks every request to a cloud model before it leaves your house: sensitive details are replaced with placeholders, the answer is translated back locally, every decision is logged. The rules come from your own policies.

Any provider with an OpenAI-compatible interface, from ChatGPT through Claude and Gemini to Mistral. Change provider and your protection stays exactly as it was.

Yes, deliberately: one discreet note per intervention, naming the policy that applies. We refuse covert filters, not least because of co-determination.

By default only metadata: time, model, detected categories, decision. Plain text requires a documented four-eyes approval, and management reports are aggregated.

Hard rules for IBANs, credit cards and secrets cannot be overruled by any AI. Uncertainty only ever errs towards caution: in doubt something is replaced or queried, never quietly let through.

Observation mode is live in one afternoon: upload the policy, confirm the proposed rules, start measuring against your own policy. Enforcement only begins once your metrics reach the agreed corridor.

SAGA is IGNAITE’s AI minute-taker. SAGA takes down meetings and committee sessions, answers questions about them and drafts the minutes, with timestamps as the source in the margin. You capture decisions with one click. At the end you decide what remains of the meeting.

To a computer in your building, and nowhere else. Recording happens in the browser of the device in the room; recognition, answers and minutes are produced on the SAGA computer. No cloud, no telemetry, no emails. The operating logs contain numbers, counters and states, never content.

SAGA separates the voices and asks who is speaking when a new voice appears. Only someone a person has named or confirmed goes into the minutes as a speaker. If members should be recognised by voice, there are voice profiles: only with signed consent, per person and committee, switched off by default. What is stored is 192 numbers, no audio.

You decide when closing the meeting. Archiving keeps the transcript and the minutes; close and delete keeps only the minutes and the header data. SAGA deletes the audio in every case. Deletion is permanent, with no recycle bin, and a deletion report records what was deleted, when and by whom.

SAGA builds on LOKI’s local speech recognition and on LOKI’s connection to local language models, both running on your hardware. You export the approved minutes as Word or PDF. Once they sit where LOKI reads, LOKI later finds the decision with its source.

No. SAGA takes the writing off their hands, the decisions stay with them: they accept or reject suggestions and approve the minutes. Answers and drafts can contain mistakes. That is why SAGA never changes anything silently.

Fixed packages with a monthly price scaled to your user count (Starter, Pro, Enterprise), plus a one-time setup including server hardware. We share the concrete figure in a first conversation, tailored to your needs. The license always covers unlimited users; operations, maintenance, support, and all updates are included. And you test risk-free: 30-day money-back guarantee.

The license itself is unlimited. Only the hardware's computing power is finite. The packages make sure your system is sized to perform for your user count. When your team grows, the hardware simply grows with it.

Yes. All product updates and new features are part of the license and applied by us. No extra costs, no project effort on your side.

Yes. Server hardware sized to your user count is included in the setup fee. Optionally, you can lease the hardware instead of buying it.

You start as a regular customer. If LOKI doesn't convince you within the first 30 days, you get your money back in full. So you test without any risk.

We handle setup, operations, maintenance, and updates of your LOKI environment, from the Pro package with proactive monitoring on top. Your IT doesn't need to build AI expertise. You simply use the product.

For the local AI server we use standard servers with current GPU accelerators: the same class of hardware data centres run their models on, sized to fit your server cabinet. We set them up at your site and handle operations & updates. A single unit is sufficient for smaller operations; if you need more capacity for concurrent queries, additional units simply stack. On the shop floor, we deploy our own MINERVA edge boards, currently in the design partner program. SAGA runs on a computer in your building, and recording happens in the browser of the device in the room. The hardware is included in the setup, belongs to you (leasing optional), and stays behind your firewall.

For operation, yes. The solution runs entirely offline. For optional support and updates we provide secure, customer-controlled remote maintenance that you can disable at any time.

We use open-weights language models selected and deployed for your specific use case. By default there are no API calls to OpenAI, Anthropic, or Google, and inference happens entirely on your hardware. You can deliberately enable cloud models in LOKI, and HEIMDALL then checks every request.

Yes. In the delivered state your data does not leave your network, there is no third-country transfer, and we are prepared for the EU AI Act. If you deliberately enable an external provider, that no longer holds for that one route. Then HEIMDALL in LOKI checks every request, replaces sensitive details with placeholders and logs every call. The on-premises architecture makes data protection technically verifiable rather than merely promised in a contract.

Yes. We integrate data sources such as document repositories (SharePoint, Nextcloud, file shares), ERP and CRM systems, email servers, and machine controls. Typically without changes to your existing infrastructure.

The path is the same every time and therefore predictable: demo, paid evaluation, first department live, rollout. After the demo you know what IGNAITE delivers for you, before you commit.

Typically a few weeks. The exact effort depends on your data landscape and the systems being connected. We accompany the entire process and only hand over when everything runs in production quality.