White Paper 01:
The integration of Generative Artificial Intelligence into the modern business ecosystem marked the start of a new era of productivity, but it brought with it a structural vulnerability that many organisations still ignore. During the first wave of adoption, the market gave in to the convenience of the public cloud, overlooking the fact that corporate data is today the most valuable asset of any institution. When a company submits its strategies, proprietary code or sensitive customer data to global language models, it is accepting a silent leak of intellectual property. In this scenario, data security stops being a merely technical matter and becomes a strategic imperative for survival.
This technological crossroads led to the identification of what we call the Modern AI Trilemma. Traditionally, companies were forced to balance three pillars that seemed irreconcilable: total security, absolute privacy and high-level performance. In conventional cloud-based architecture, security and privacy are invariably sacrificed in exchange for scalable processing capacity. On the other hand, earlier attempts at local processing often fell short on performance, resulting in slow systems unable to handle the complexity of world-class language models.
SOIA emerges to break this trilemma by implementing a local, native AI architecture. We believe that true digital sovereignty is only achieved when intelligence lives where the data is born: within the company's own digital walls. By using AI-certified, high-density computing equipment, we bring the power of a data centre to an edge infrastructure (Edge AI) that operates 100% offline. This approach ensures that intranet processing preserves the integrity of trade secrets, while local information libraries allow AI agents to collaborate with the human team with virtually no latency.
Ultimately, the path to the future does not lie in dependence on external servers or generic models exposed to global threats. The future demands a transition to sovereign systems, where each organisation owns its own intelligence and infrastructure. By integrating digital co-workers that respect the company's physical perimeter, we turn technology from a potential vulnerability into an unquestionably powerful tool. The era of offline AI is not just an alternative; it is the foundation on which visionary companies will build their leadership in the 21st century.
Offline AI Architecture and the Corporate Security Trilemma
Native Architecture and Digital Sovereignty, or SOIA's "Trust Engineering"
True innovation in enterprise Artificial Intelligence lies not only in the language model, but in the architecture that supports it. SOIA stands out as a platform built in native code, designed to operate as an intelligent resource orchestrator. Unlike generic solutions, SOIA dynamically manages multiple libraries, tools and APIs, selecting the most suitable computing resources based on the complexity of the task and the performance needs of each agent.
This engineering enables unprecedented vertical and horizontal scalability in controlled environments. Whether the platform is running on an isolated intranet or connected to the network, its core engine decides in milliseconds which technical "muscles" to activate. This modularity ensures the platform is fully tailored to each customer's workflow, adapting from small service operations to complex industrial or government ecosystems.
Security is implemented through overlapping layers, or Security Gates, which guarantee the full integrity of the system. SOIA's structure is organised into personalised spaces for the different human employees; each user works in an isolated, secure environment, accessing only the agents and databases specific to their role. This compartmentalisation of privileges ensures that sensitive information never moves between departments without proper authorisation.
To deepen the technical understanding of our infrastructure, we look at the pillars of the platform:
Q: What does it mean, in practice, that SOIA uses "native code" to manage libraries and APIs?
A: It means SOIA is not just an intermediary for external models. We have a proprietary engine that intelligently manages assets. If an agent needs to perform a complex calculation, the platform activates a local maths library; if it needs an external market analysis (when connected), it calls the specific API. This orchestration ensures performance is optimised and the agent always has the right tool for the job, with no wasted resources.
Q: How is security guaranteed in an environment where multiple employees interact with the AI?
A: Through an architecture of "Personalised Spaces" and multiple Security Gates. Each employee has their own work silo. The agent that supports Human Resources is not the same, nor does it have access to the same data, as the agent that supports the Engineering team. These security gates prevent unwanted data crossover and ensure information sovereignty is maintained at every level of the hierarchy.
Q: Is the platform scalable within a limited local infrastructure?
A: Yes. Because it is a tailored platform, SOIA is configured to grow with the company's needs. Our native code allows efficient memory and processing management, making it possible to add new agents and expand the knowledge libraries without restructuring the customer's entire network. It is a solution built to measure that scales with the business.
Q: What is the advantage of having specific agents for each employee instead of a centralised AI?
A: Specificity creates precision. A generalist agent makes mistakes because of too much irrelevant information. In SOIA, agents are trained and equipped only with the tools needed for the user's role. This not only increases productivity but also strengthens security, since each agent's scope of action is strictly limited by its original purpose.
SOIA - Organisational Artificial Intelligence System Private artificial intelligence | Human collaboration
SOIA Interface:
SOIA: A new way for Humans and Artificial Intelligence to work together
SOIA is not just another AI tool; it is a work ecosystem designed to seamlessly integrate intelligent agents and human teams. In a world where efficiency and data security are critical priorities, our platform offers a unique and exclusive solution.
SOIA's great differentiator is its flexibility. We work with a hybrid model: the platform runs Offline, but it can also run on an Intranet or with specific online modules (e.g. an email client or a specific API), ensuring that the processing of sensitive information stays within the company's servers and never leaves the local network. The customer chooses their own security boundary. Through information libraries stored on the computer or server itself, our agents keep their full ability to respond, even without an internet connection.
Every business is unique, and SOIA reflects that. The platform is fully customisable, allowing the customer to choose and shape the agents and functions that best fit their workflows, from industry to services, from the public to the private sector.
In the following pages, we offer a guided tour of our interface, detailing the menus and features that make this symbiosis possible.
Offline AI Architecture and the SOIA-AI User Interface
Image 6: Interface customisation option
Workspace customisation lets you adapt the SOIA interface to each organisation's preferences and context of use.
The user can choose between:
light or dark mode, according to visual preferences and working conditions;
interface language, with the option to switch between Portuguese and English;
operating environment, showing whether SOIA is running locally, on an intranet or in another configured context.
This customisation improves the user experience without changing the agents' functions or the work structure. The interface keeps the same menus, agents, directives and knowledge, but adapts the presentation and operating environment to each team's needs.
Image 5: Team menu
Team is the menu for managing the people who have access to the SOIA space. It lets you view existing members, control access permissions and invite new colleagues to join the workspace.
Human collaboration is integrated with the AI agents, allowing members of the organisation to work on the same context, knowledge and set of tools. From this menu you can:
view the members with access to the Workspace;
check their addresses and invitation status;
prepare and send invitations to new colleagues;
manage people's access to the SOIA environment;
ensure the platform is used in a controlled way.
In this way, SOIA is not just a set of isolated agents: it combines AI agents, shared information and collaboration between team members, while keeping access management centralised.
Image 4: Brain menu
Brain is SOIA's local knowledge repository. It lets you store, organise and make information available to the AI agents, keeping knowledge separated according to each agent's role.
Knowledge can be organised by agent:
Raquel: information related to secretarial work, scheduling and organisation;
Luis: executive documentation and decision support;
SOIA: general knowledge of the platform, coordination and institutional context.
The Brain supports different forms of knowledge:
Memories: notes, priorities, instructions or learned information;
Images: visual elements, diagrams, screenshots or other graphic resources;
Pages: content from web pages;
Documents: files such as PDF, DOCX or other supported formats;
All: a combined view of all available content.
This organisation lets each agent consult only the knowledge relevant to its role, reducing the mixing of contexts and making answers more accurate and controlled. Through the Add knowledge option, the user can feed the Brain with new information and keep the agents up to date.
Image 3: Workspace menu
Workspace is the space where you define the organisation's shared context and the directives that guide all of SOIA's AI agents.
In this menu you can configure:
the company's description and activity;
the target audience and operating context;
identification, sector of activity and country;
the language of communication;
the tone of voice the agents should use.
This information is shared with the active agents, allowing them to respond in line with the organisation's reality, goals and identity. Each agent keeps its specific role, but works from a common base of context and guidance.
Once the changes are made, the directives can be saved and applied to the workspace.
Image 2: Agents space
Agents is the menu for managing and interacting with SOIA's AI agents. It lets you see the active agents, learn each one's role and start a direct conversation with the selected agent.
In this space, each agent takes on a specific role:
Raquel, as AI secretary, organises information, schedules and next steps;
Luis, as executive agent, supports management and decision-making tasks;
SOIA, as coordination agent, links information and activities between the other agents.
The user can make requests, ask for analyses, set priorities or ask for notes and tasks to be prepared. Answers are shown in a contextual conversation and may include references to the Workspace, the Brain or other available resources.
This menu therefore lets you use the agents individually and, in future, coordinate collaboration between several specialised agents within the same workspace.
Image 1: AI Environment menu, Light Mode
AI Environment is the central dashboard of the SOIA workspace. It gives an overview of the organisation's activity and lets you follow, in one place:
the active AI agents and their roles;
the directives and rules applied to the environment;
the documents, pages and memories available in the Brain;
the team members with access to the platform;
recent activity and changes made.
From this menu, users can also quickly start actions such as creating a new agent, setting a directive, adding knowledge to the Brain or inviting colleagues. It works as the entry point for coordinating people, AI agents and internal information in a controlled environment.
Want to try the platform live on our website?
White Paper 02:
The demo shows an early version of the interface, with no documents loaded, so it does not answer questions.
SOIA structure:
SOIA flowchart:
SOIA organisation charts:
SOIA is an ecosystem in constant improvement and evolution, adapting to the constraints of the ecosystem it is developed in, so the following information may have changed slightly. Even so, this is the general footprint of SOIA's path and flow of information.
1. Technical Chart (Code and Network Structure)
This diagram shows how the system is divided into modules (isolated containers) that communicate with each other through a secure internal network, with no access to the external internet.
Interface (Frontend): The user accesses it locally (localhost:8000). The interface loads the history, sends the questions and displays the answers with buttons and sources.
API Gateway & Orchestrator: This is the central control point (api_gateway). It receives the request, validates the paths, performs the security screening (like "step 1" of the previous flowchart) and routes it to the supervisor.
Agent Runtime (Agents): This is where the agent's behaviour is shaped through profiles and directives (prompts). This module connects to the local AI models.
Local Models (Ollama): The system runs local language models on the GPU/processor (such as qwen3:8b for text and qwen3-vl:8b for images) and a vector model (bge-m3) to search documents.
Tools Service: An isolated module to handle heavy tasks, such as file conversion (Word, Excel, PDF), audits and data ingestion into the "Brain".
Data (Storage folder): Where the company's general documents, approvals, drafts and audits are stored securely.
2. Functional Chart (Workflow and Profiles)
This diagram focuses on business rules and on how the agents behave towards the Human user, who always has the final say on sensitive actions.
The Human at the Top: The user talks to the agents, uploads documents, sets rules and approves or rejects pending actions.
Approval Mechanism:
Without approval: General queries to the Brain, translating answers or generating temporary Word/Excel drafts happen instantly.
With approval (Approvals board): Saving/deleting data in the Brain, handling GDPR matters or generating final documents is frozen until the human approves.
The Three Agent Profiles: The system splits tasks into three areas of expertise, each with its own "private Brain" and isolated history:
Founder Associate: Focused on strategy, decisions, priorities and investors.
Data Partner: Focused on data, KPIs, numbers and audits (due diligence).
HR & People: Focused on human resources, teams and internal organisation.
Handover: If the user asks the strategy agent a human resources question, the Supervisor detects the mismatch and proposes to the human that the conversation be passed to the right agent, transferring only the necessary context.
The system processes users' questions sequentially through 4 main decision layers:
1. Screening and Security: Immediately blocks requests for credentials or personal data (fixed refusal response) and translates the text for search if it is in another language.
2. Supervisor (Business Rules): Detects sensitive actions that require human approval (such as GDPR or payments), handles greetings and simple calculations, and triggers attachment analysis if there are files in the window.
3. Brain Layer (Data): Activated only for complex queries. Uses specific tools (brain_search or analyse_tables) to look for data and evidence in the company's documents.
4. Courtesy and General Knowledge: Answers opinions or general knowledge questions based on the history and the language model's knowledge.
Finalisation (Post-Decision): Before delivering the answer to the interface, the system uses Python code to ensure the use of European Portuguese (formal register), protect the internal instructions and record the interaction in the log.
SOIA
Private, offline AI for companies.
© 2026 SOIA. All rights reserved.
