Document search
Find a contract, a procedure or a specific version without knowing which folder it sits in.
Cloud and datacenter · Private artificial intelligence
Private AI: an assistant for your company's documents and data
We bring your information together on an AI and mass-storage platform running on dedicated hardware, with a conversational interface for finding information, summarising documents and producing statistics.
- Dedicated hardware
- Up to 200 TB raw
- 128 GB of unified memory
- Optional private network up to 10 Gb/s
What Private AI is and what it depends on
Private AI is a cloud AI and mass-storage solution running on dedicated hardware: a machine built around an AMD Ryzen AI Max+ 395 processor, 128 GB of unified memory and up to 200 TB raw, used through a conversational interface to query company documents and data.
The answers depend on the data, sources and integrations configured for each project. We do not promise connectors or features that are not implemented within your scope.
Use cases
What an assistant over your own information is useful for
A private assistant answers questions from the documents and data your company provides. The value depends on the quality and organisation of that information, not on the platform on its own.
Internal knowledge
Ask about criteria, resolved incidents or organisational policies in plain language.
Summaries
Get a summary of long minutes, reports or files, with access to the original document.
Information analysis
Compare texts or pull the key points out of a set of documents.
Statistics
Produce counts and trends from the data that has been connected and validated.
Each use case requires the data, sources and integrations to be configured and tested for that project. Connectors that are not implemented are not advertised.
Privacy and control
What you control and what depends on how it is configured
Hosting AI and data on private dedicated infrastructure gives you more control over the environment: centralised storage, permissions defined per organisation and the ability to limit the exposure of sensitive information.
That privacy is not automatic: it depends on whether the flow uses local models or external services.
What the private environment brings
- Control of the environment where models run and data is stored.
- Centralised storage instead of scattered copies.
- Permissions defined per organisation.
- Less exposure of sensitive information to third-party services.
What has to be verified
- Whether a query uses a local model or an external API.
- Whether the whole flow (indexing, embeddings, answers) stays inside the environment.
- The processing terms of any third-party integration.
We only state that queries and documents stay inside the private environment when the configuration uses local models and that has been verified across the whole flow. With external services or third-party integrations, processing depends on their terms. We do not present the service as 100% secure, nor as physically isolated for each client, nor as guaranteeing GDPR compliance on its own.
Technical specification
Configurable hardware and storage
The platform is sized to the volume of information and the models that will run. Storage combines NVMe and HDD, each with a different role.
- NVMe for frequently accessed data, models or cache.
- HDD for large volumes and archive.
- Up to 200 TB raw: up to 160 TB HDD and 40 TB NVMe, depending on the disks installed.
The 200 TB figure is maximum raw capacity. Usable space will depend on the disks chosen, the format, the redundancy configuration and the file system.
Integration
Connected to your cloud or dedicated servers
The solution can optionally connect to Seintec cloud or dedicated servers over a private network of up to 10 Gb/s, depending on the architecture and connectivity contracted.
It can also integrate externalised backups.
The private network and externalised backups are configurable services, not features automatically included with the platform.
Service scope
What we define with you and what is left out
The proposal puts in writing what the service includes and what is configured separately, so nothing has to be assumed.
Defined in the project
- Hardware, disks and capacity sized to volume and workloads.
- Which models run and how they are accessed.
- Which sources, permissions and integrations are connected.
- Administration, monitoring and support per agreed model.
Not included by default
- Connectors to applications that have not been contracted.
- Additional encryption, two-factor authentication, auditing or retention.
- Certifications or accredited legal compliance.
- Physical isolation dedicated to a single company.
If you need additional isolation, that is a design and scope decision: it is assessed and quoted separately.
Several companies on the same hardware
Separate sessions, credentials and workspaces
When the architecture serves several companies on the same hardware, we do not share sessions, credentials or workspaces between organisations.
Separation is described only in the terms the implementation allows you to verify; if stronger isolation is required, it is raised as a design and scope decision.
How it gets started
- 01
Data and use-case review
We inventory documents and systems and pick the use cases that deliver value. Without organised data, the assistant does not answer well.
- 02
Sizing
We choose disks, memory and models based on volume, concurrency and the type of queries. Raw capacity is not usable space.
- 03
Deployment
We prepare the platform, install the disks and leave the environment ready to load the sources.
- 04
Sources, permissions and integrations
We connect the agreed sources, define permissions per organisation and test that each user sees what they should.
- 05
Testing and handover
We validate answers against your documents, document the scope and agree how the service is run.
Timeline and responsibilities are agreed in the proposal. We do not promise universal deadlines or a go-live date without reviewing the project.
Platform specifications
Maximum capabilities of the platform. What is installed and what is contracted depends on your project.
| Component | Capability or condition |
|---|---|
| Processor | AMD Ryzen AI Max+ 395 |
| Unified memory | 128 GB LPDDR5X |
| Compute capacity | Up to 126 TOPS of total compute capacity; not an NPU-only figure |
| Disk bays | Five bays for HDD SATA disks |
| Expansion slots | Five M.2 NVMe slots |
| Raw storage | Up to 200 TB raw: up to 160 TB HDD and up to 40 TB NVMe, depending on the disks installed |
| Recommended NVMe use | Frequently accessed data, models or cache |
| Recommended HDD use | Large volumes and archive |
| Private network (optional) | Up to 10 Gb/s to Seintec cloud or dedicated servers, depending on the architecture and connectivity contracted |
| Backups (optional) | Externalised, as a configurable service |
Frequently asked questions about Private AI
Do the data and documents leave the private environment?
It depends on the configuration. If the whole flow uses local models and that is verified, queries and documents stay inside the environment. If external models or APIs, or third-party integrations, are used, processing depends on their terms.
Are the 200 TB usable space?
No. That is maximum raw capacity. Usable space depends on the disks chosen, the format, the redundancy configuration and the file system.
Are the 126 TOPS the NPU's?
No. They are the platform's total compute capacity and should not be described as NPU-only power.
Can I query everything without configuring anything?
No. Each use case requires the data, sources and integrations to be configured and tested. Without connected sources there are no reliable answers.
Is the 10 Gb/s private network included?
Not automatically. It is a configurable option, alongside externalised backups, depending on the architecture and connectivity contracted.
Does it meet GDPR on its own?
We do not claim that. Compliance depends on the processing, the purposes, the contracts with processors and the measures agreed; the platform does not guarantee it by itself.
Can I have hardware isolated just for me?
It can be raised as a design and scope decision and is assessed separately. By default we do not present the service as physically isolated for each client.
Next step
Let's review your data and your use case
Tell us which documents and data you want to query, with how many people and which integrations you need. We will propose hardware, storage and scope.