Deployment options¶
zGentic can be used in four ways. They differ in who runs the platform, where it runs and how much help you get. The product is the same in all four. Any of them can use AI models that you run yourself, see Your own models below.
| 1. SaaS | 2. SaaS with professional support | 3. Dedicated SaaS instance | 4. Your own environment | |
|---|---|---|---|---|
| Who runs the platform | zGentic | zGentic | zGentic | You (with our help, if you want it) |
| Where it runs | zGentic's shared cloud, EU (Frankfurt) | zGentic's shared cloud, EU (Frankfurt) | A separate instance for you alone, in the region you agree with us | Your data centre or your own cloud account |
| Separation from other customers | Each organisation is fully separated within the shared platform | Same as 1 | A separate instance: its own database, storage and workloads | Entirely yours |
| Updates | Automatic | Automatic | Scheduled with you | You decide when to install a release |
| Professional support | Not included: documentation and the in-app feedback channel | Included | Included | Available |
| Your own AI models | Yes | Yes | Yes | Yes |
1. SaaS without professional support¶
Sign up, invite your colleagues and start. zGentic runs and updates the platform. Your organisation is fully separated from every other: answers only ever use material your people are allowed to open, and nothing is shared across organisations.
Help comes from this documentation and from the feedback button in the app. There is no named contact and no agreed response time.
Good for: teams that want to start quickly and can find their own way.
2. SaaS with professional support¶
The same shared platform as option 1, plus a support agreement. Typically that covers:
- help with setting up your organisation: identity and sign-in, model providers, connectors, approvals and roles;
- guidance on designing agents and on governance, for example preparing for ISO/IEC 42001;
- a named contact and an escalation path, with response times set out in your agreement.
Good for: organisations that rely on zGentic for real work and want someone to call.
3. SaaS as a dedicated instance¶
zGentic runs a separate instance only for you, with professional support included. Nothing is shared with other customers: not the database, not the file storage and not the running services. Region, maintenance windows and release timing are agreed with you, and your identity provider can be connected directly.
Good for: regulated or security-sensitive organisations that want SaaS convenience without sharing infrastructure, or that need a specific region.
4. Deployment in your own environment¶
You run zGentic inside your own infrastructure: on-premises or in a cloud account you control. Your data stays within your environment. zGentic provides the release packages and, if you want it, professional support for installation, upgrades and operation.
What your environment needs, in short:
- a Kubernetes cluster (three nodes as a starting point), with a network layer that enforces network policies;
- block storage for the database, and S3-compatible object storage for files and backups;
- gVisor on the nodes, which isolates the code the assistant writes and runs;
- a private container registry holding the zGentic images;
- your identity provider (OIDC or SAML 2.0) for sign-in;
- a secrets store and certificates for the web address you will use;
- outbound internet access for the model providers, web search and connectors you choose. If you use only your own models and internal connectors, outbound access can be limited to what those need.
Sizing depends on how many people, documents and concurrent tasks you expect; we size it with you.
Good for: organisations whose policies require that the platform and its data stay in their own environment.
Your own models¶
In every option you can use models that you host yourself, for example open-weight models served by your own inference server, alongside or instead of commercial model providers.
- In Admin → Model backend, add a provider of type OpenAI (custom endpoint) (or LiteLLM) and enter your server's address and key. Your server needs to offer an OpenAI-compatible API, which is what common inference servers provide.
- In Model settings, choose which of your models people may use, which one is the default, and the fallback order. You can mix your own models with commercial ones, for example your own model as the default and a commercial model as the fallback.
- Model choice is per organisation, so in a shared deployment your models are never offered to anyone else.
Where the model runs makes a difference:
- Options 1–3 (SaaS): your inference server must be reachable from zGentic over a secured connection. Your prompts and answers travel between zGentic and your server, and the model itself never leaves your infrastructure.
- Option 4 (your environment): your inference server runs next to zGentic in your network, so nothing leaves your environment for the model step.
zGentic itself does not host the models. You run the inference server on your own hardware, GPUs or GPU cloud, and zGentic connects to it like to any other provider.
Choosing¶
- Want to start today? Option 1.
- Using it for important work and want a contact? Option 2.
- Need your own infrastructure but not the operations work? Option 3.
- Must keep everything in your own environment? Option 4.
For options 2–4, contact zGentic to agree on scope and terms.