Model as a Service · NapsixAI
Your organization.
Your intelligence.
Your terms.
We design, build and operate artificial intelligence systems adapted to your data, processes and security requirements.
From research to operations, on dedicated infrastructure, in your cloud or on your servers, depending on the project.
An architecture designed around your organization. Conceptual illustration.
01 / The service
A complete view.
A solution built for you.
There is no single way to build enterprise AI. We design every layer around the problem, the context and the people who will use it.
Research and design
We define the problem, evaluate models and design an architecture around your goals and constraints.
Data and knowledge
We prepare sources, permissions and knowledge bases so the AI works with relevant information.
Model adaptation
We combine instructions, retrieval and fine-tuning when evaluation shows it adds value.
Applications and agents
We connect interfaces, tools and systems with permissions that match each task and user.
Infrastructure and inference
We design the compute environment and model execution to meet the needs of the operation.
Security and evolution
We evaluate versions, define controls and support monitoring and continuous improvement.
02 / Architecture and deployment
Infrastructure adapts to your rules.
From a solution operated by Napsix to your own servers. We evaluate the right option for your requirements, resources and team.
Dedicated infrastructure
An environment for your project, operated by Napsix under the agreed service.
Your cloud
Deployment in your organization's cloud account, with controls defined together with your team.
Your servers
We evaluate deployment in your data center, with a compatible model and hardware.
Select an environment
// SELECT AN ENVIRONMENT
Napsix operates the environment
The model and its applications are deployed on infrastructure dedicated to the project, with agreed responsibilities and controls.
Illustrative diagram. Architecture and controls are defined for each project.
Feasibility depends on the model license, hardware, connectivity and required components. An isolated network requires a specific evaluation.
03 / Security and control
Security at every layer.
Peace of mind is built with concrete decisions. We define controls from design and responsibilities throughout operations.
Data under control
Classification, access, residency, retention and authorized use for training, defined for each project.
Infrastructure and access
Isolation, network, encryption, identities and permissions according to the environment and agreed requirements.
Models and agents
Evaluations, versions and tool permissions, with human review for sensitive actions where appropriate.
Operations and response
Monitoring, logs, updates and incident response responsibilities defined in the scope.
Contractual confidentiality
We can formalize an NDA before sharing sensitive information. Depending on the project, exclusivity or non-compete terms can also be agreed, with defined scope and duration.
Clear rights and exit
Your data remains yours. Rights over developments, the adapted model and transfer are agreed by contract, respecting third-party licenses.
Controls, services and responsibilities are established according to each project's architecture and contract.
04 / Technical validation
We do not outsource it.
We build the model.
Mato v1.0 is the proof: it starts from Qwen 3.8 27B by Alibaba Cloud, was trained on anonymized Napsix.com records and runs on NVIDIA H100 in Daytona. The weights are published on Hugging Face.
256k
native context
26
native apps
17.7×
concurrency
Measured in production (2026-09) on Daytona H100. Varies with load.
Alibaba Cloud
Qwen 3.8 27B base · Apache-2.0
Daytona
NVIDIA H100 80GB · Startup Grid
NVIDIA Inception
NapsixB2B · program member
Hugging Face
Open weights · Mato v1.0
Other technologies
We also evaluate
AnthropicManaged models
Google CloudInfrastructure and services
Gemma 4Google model family
DigitalOceanCloud infrastructure
AWSInfrastructure and servicesModels with available weights. We can evaluate alternatives such as Qwen or Gemma for adaptation and deployment on controlled infrastructure, subject to their license.
Managed services. OpenAI, Anthropic and other providers can be integrated by API when the project's policies allow it.
Brands identify technologies and providers that may be evaluated or used; they do not imply the same kind of commercial relationship. Model origin is one criterion among several, along with license, performance and security.
NVIDIA, the NVIDIA logo and NVIDIA Inception are trademarks of NVIDIA Corporation. NapsixB2B is a member of NVIDIA Inception. Membership does not imply NVIDIA's endorsement of our products or services. Mato v1.0 is distributed under the Apache-2.0 license; the weights are published on Hugging Face.
05 / Applied experience
Meet Mato.
Our first model.
Before offering this path, we walked it with our own ecosystem.
Mato v1.0 starts from Qwen 3.8 27B by Alibaba Cloud and runs on NVIDIA H100 in Daytona. We trained it on anonymized Napsix.com records and published it on Hugging Face.
Mato is our own, open case. An enterprise project may have different privacy, licensing and distribution conditions.
Alibaba CloudDaytonaNVIDIA InceptionHugging FaceFrom a base model to a solution for our ecosystem.
06 / How we work
From a question to a solution in production.
We move in stages, with shared goals and decisions. The depth of each phase depends on the project.
Confidential diagnosis
Goals, use case, data and constraints. We agree on what a good result would be.
Research and architecture
We evaluate models, infrastructure, security and costs to propose a viable path.
Controlled test
We test with agreed cases, measure results and decide how to continue.
Development and adaptation
We prepare data, connect systems and adapt the model when needed.
Validation and deployment
We test the solution and put it live in the environment defined with your team.
Continuous operations
We monitor, evaluate new versions and improve within the agreed scope.
07 / Applications
Intelligence starts
in your context.
Every organization has different knowledge and responsibilities. These are some starting points for designing a solution.
Companies and corporations
Assist teams, connect knowledge and support processes across areas and systems.
Agencies and the public sector
Explore document and operational assistance with defined control and procurement requirements.
Specialized operations
Apply technical knowledge to specific tasks, with permissions and review matched to the risk.
Illustrative scenarios; scope is evaluated for each organization.
08 / Frequently asked questions
Clarity is also part of the service.
Do we need to train our own model?
Not always. To query up-to-date documentation, a knowledge base can be enough. We evaluate instructions, retrieval and fine-tuning based on the task and the results.
What data leaves to external providers?
It depends on the architecture. We define what information each component may process and whether external services are allowed. A model connected by API may require sending data to the corresponding provider.
Can we choose where the model comes from?
Yes. We evaluate models from different origins, including US and Chinese providers, based on performance, license, infrastructure, security and organizational requirements. Origin is one of several selection criteria.
Can it run on our servers?
We can evaluate deployment in your cloud or data center for compatible models. Feasibility depends on the license, hardware, connectivity and required components. An isolated network needs a specific evaluation.
Who owns the model and the developments?
Your data remains yours. Rights over the adapted model, code, documentation and transfer are defined in the contract, respecting the base model license and third-party components.
Do we sign a confidentiality agreement?
We can formalize an NDA before sharing sensitive information. Depending on the project, exclusivity or non-compete terms may also be agreed, with defined scope and duration.
How do we verify it works?
We agree on cases and evaluation criteria with your team. Versions are compared before deployment and review and improvement processes are defined according to the use case risk.
How much does it cost and how long does it take?
It depends on the use case, the data, the infrastructure and the service level. The diagnosis defines the scope and prepares a phased proposal, including implementation and operations costs.
The next step
Let's talk about the intelligence your organization needs.
A conversation to understand your goals, requirements and options. We can formalize confidentiality before sharing sensitive information.
You pick the time in the NapsixAI calendar.