Application installation
On a local computer or a server, after a support and maintenance agreement is signed.
FOR BUSINESS, RESEARCH AND ENGINEERING
One goal.An orchestra of AI models.
A universal autonomous multi-agent platform.
Bring different models, agents and tools together for business process automation, research and AI engineering.
06orchestration modes
19agent instruction templates
MCPyour tools and services
CLEAR TERMS FROM THE FIRST STEP
Your infrastructure.
Our support.
Start with a task, choose an environment and agree the service scope. Understand the costs before installation.
Maintenance and support
Service scope, connections and support terms are agreed individually. Free installation does not remove support costs.
Model tokens and APIs
The client pays for model usage under the chosen providers' rates and manages their own limits.
Where will your AI team run?
We check server OS, resources and executor compatibility. Cloud model requests are sent to the respective model providers.
CONTROL THE PROCESS
Fewer disconnected actions.
A more coherent workflow.
Define the goal, participants and rules. Build a repeatable process and assess its output against your criteria.
Match models to the task
Multiple models per agent and six orchestration modes. Compare alternatives or separate drafting from review.
Explore orchestration ↘Keep the method and context
Roles, projects, instructions, memory and MCP in one workspace. Research methods, TRIZ and first principles structure analysis.
Choose an approach ↘Repeat an agreed process
Teams share a goal and run manually, at a scheduled time or on an interval. Scheduling requires a running core.
Explore automation ↘01 / WORKSPACE
Models organised
around your task.
Configure agents, give them a shared goal and decide how results are prepared and reviewed. Projects, conversations and tools stay in one workspace.
Agents with clear roles
Researcher, analyst, developer or a custom role. Each has instructions, models, skills and selected connections.
Multiple models per agent
Select connected models, set their order and choose a coordinator for the task.
Project context
Keep conversations, materials and useful knowledge connected to the project.
02 / MODEL ORCHESTRA
Choose the participants.
And how they collaborate.
Independent alternatives, a sequential chain or a dedicated reviewer. The mode defines how your selected models work together.
Orchestra members
This website demonstrates the configuration. No model requests are sent.
TASK
Coordinator output
Independent answers and a combined result
Models prepare alternatives independently. The coordinator compares them and produces the final output.
The app supports any number of configured models. Four examples are shown here. API orchestra participants primarily handle text; authorised tools run through the main agent. Each call can consume provider limits.
03 / WORKING METHODS
A research question.
An engineering approach.
Choose instructions that structure the work: hypotheses and sources, constraints, calculations and verification.
Research
Questions, sources, alternative hypotheses and attempts to disprove them.
Systems and first principles
Connections, system boundaries, assumptions and constraints.
TRIZ and custom methods
Working with contradictions and your own domain-specific instructions.
Methods are instruction templates. Scientific conclusions and calculations require independent verification.
04 / APPLICATIONS
One platform.
Different classes of task.
01 / BUSINESS PROCESSES
Repeatable work with a shared goal.
Analytics, regular reports, documents and materials. A team works from one brief, and a lead combines the output.
Example process
- Task and acceptance criteria
- Independent preparation
- Review of assumptions and data
- Team lead's final output
CRM, ERP and other services require separate API/MCP connections and permissions.
05 / TEAMS AND AUTOMATION
One goal.
A managed process.
Set a lead, roles and acceptance criteria. Choose sequential preparation, parallel work or work with review.
- Manual
- Start when the task is ready.
- Calendar
- A date, time and time zone.
- Interval
- A repeating interval of at least five minutes or a cron rule.
The computer and core must be running. Repeated execution does not automatically verify that the goal was achieved.
06 / KNOWLEDGE AND EXTENSIONS
A workspace
with your context.
Connect tools, develop agent instructions and choose which capabilities each team member can use.
Third-party and custom tools
Configure MCP servers, including your own. Authorisation, permissions and executor compatibility are checked for each connection.
Memory and skills
Edit project memory, import text and code, and add instructions. Semantic retrieval is disabled in the reviewed version. Full RAG and evaluated self-improvement are development directions.
Images and video
Select generation providers and models, grant agent permissions and save media results. Suitable keys, provider plans and model capabilities are required.
DEPLOYMENT FOR YOUR TASK
Start with a measurable pilot.
One scenario, clear criteria and a checked result. Agree the service scope and environment before deployment.
Define the process
Goal, inputs, constraints, access and acceptance criteria.
Agree the configuration
Models, agents, orchestration, integrations and support agreement. Then install the app free of charge.
Verify in practice
Compare output to the criteria, assess costs and errors, and decide what can run regularly.
PRODUCT MATERIALS
Capabilities.
Settings. Real screens.
The existing 34-slide presentation remains available in Russian. Website content is available in English.
07 / SEE THE PLATFORM
NC AgentOS in action.
08 / QUESTIONS
Before your
first workflow.
How do I get free installation?
Send an enquiry, agree the environment and sign a support and maintenance agreement. The app is then installed free of charge locally or on a server. Support terms are agreed separately. You pay model token and API costs.
Is this a computer operating system or an AI model?
NC AgentOS is an application-level multi-agent AI operating system: a workspace for agents, models, tools and workflows. It does not replace your computer's operating system or create its own foundation model.
Can one agent use multiple models?
Yes. Select connected models, set a coordinator and choose from six orchestration modes. Actual availability depends on the selected providers.
What do agent working methods do?
They provide instruction templates, including research, systems thinking, TRIZ and first principles. They do not change a model's hidden reasoning and do not guarantee correct conclusions.
Are RAG and self-learning ready?
File memory and text import are available. Semantic RAG is disabled in the reviewed version. Full RAG and an evaluated improvement cycle require further development; model-weight training is not advertised as available.
Does scheduling work while my computer is off?
No. The core and its host must be running at the scheduled time. A server deployment requires an agreed compatible environment and operational checks.
Does local installation keep all data local?
Not automatically. When using cloud models, requests are sent to the selected providers. Deployment location, tool access and provider data policies must be evaluated separately.
PRODUCT FACTS
NC AgentOS
at a glance.
- What it is
- An application-level multi-agent AI operating system for managing agents, models, tools and workflows.
- Who it serves
- Business teams, researchers and engineering groups.
- How to get it
- Free installation locally or on a server after a support and maintenance agreement. The client pays model token and API costs.
- What needs separate verification
- Server compatibility, external integrations and result quality. Full RAG and evaluated self-improvement are development directions.
- Support team
- IRT LLC · info@irtit.ru · +7 977 484-74-68. Updated 17 October 2026.
FROM YOUR GOAL TO A WORKING PROCESS
Build an AI team
around your task.
Discuss a business process, research workflow or engineering scenario.
Discuss deployment ↗