Agents Under the Kernel
Embedded Files
Let agents provide intelligence without making them the foundation of your system.
Agents evolve. The Kernel remains.
Agents Under the Kernel
Agents Under the Kernel
Let AI agents evolve without rebuilding your system
Let AI agents evolve without rebuilding your system
AI agents are becoming a new way of building and operating software.
New agent frameworks appear constantly. New orchestration platforms emerge. New models, tools and automation engines replace older ones.
Your business system should not have to change every time your AI stack does.
HexaKernel introduces a simple architectural principle:
Agents belong below the Kernel.
Agents belong below the Kernel.
Agents, workflow engines, automation platforms and AI services become execution components of the system — not the system itself.
The domain, identity, security, governance, observability and execution model remain above them, in a stable layer provided by HexaKernel.
A stable Kernel. An evolving agent layer.
A stable Kernel. An evolving agent layer.
Instead of building your application around a particular agent platform, HexaKernel provides a stable domain and execution foundation.
The components underneath can evolve.
The Kernel remains.
One identity across everything
One identity across everything
Every agent, workflow, service and human interaction operates within the same application identity model.
No fragmented authentication.
No separate security model for every agent platform.
No need to reconstruct who did what across multiple systems.
HexaKernel provides a unified foundation for:
Identity
Authentication
Authorization
Roles
Permissions
Tenancy
Context
Service-to-service access
An agent can act on behalf of a user, a role or a business process while remaining inside the same security model as the rest of the application.
One system. One identity model.
One system. One identity model.
Unified observability
Unified observability
When agents operate across multiple platforms, understanding what actually happened can become difficult.
A business operation might involve:
Without a common execution layer, every component can produce its own logs, traces and metrics.
HexaKernel provides a common execution context.
That makes it possible to observe the complete operation as one business transaction, rather than as a collection of unrelated technical events.
You can understand:
What happened
Who initiated it
Which agent acted
Which tools were used
Which domain operations were executed
Which decisions were made
Which policies were applied
Which systems were affected
Where the process failed
How long each step took
Observe the business operation, not just the individual services.
Observe the business operation, not just the individual services.
Unified auditability
Unified auditability
AI introduces a new requirement:
You need to know not only what the system did, but why and under which authority it did it.
HexaKernel provides a unified audit foundation for human and machine activity.
Whether an action originates from:
a human user,
an AI agent,
an automated workflow,
an API,
or another system,
it can participate in the same domain-level audit model.
The result is a continuous chain -->
AI activity becomes part of the application's audit trail — not a separate universe.
AI activity becomes part of the application's audit trail — not a separate universe.
Agents become replaceable
Agents become replaceable
This is one of the most important consequences of the architecture.
Suppose today you use one workflow engine.
Tomorrow, a better platform appears.
You should not have to redesign your application.
The architecture becomes -->
The underlying agent technology can change without changing the fundamental domain model.
Replace the agent. Keep the system.
Replace the agent. Keep the system.
This makes technological migration dramatically easier.
If a new and better orchestration platform replaces your current one, the migration becomes an implementation concern rather than a domain rewrite.
Technology becomes an implementation detail
Technology becomes an implementation detail
The same principle applies beyond agents.
Your business should not be permanently coupled to:
one AI model provider,
one workflow engine,
one agent framework,
one vector database,
one integration platform,
one infrastructure technology,
or one generation of AI tooling.
The domain remains stable.
The implementation underneath can evolve -->
The business does not need to change simply because technology does.
Integration becomes a capability, not a collection of hacks
Integration becomes a capability, not a collection of hacks
Complex systems rarely live in isolation.
They interact with:
CRMs
ERPs
document systems
payment platforms
communication tools
government services
external APIs
data providers
AI services
internal applications
When these integrations are built directly into individual agents and workflows, the architecture becomes increasingly fragmented.
HexaKernel provides a common domain boundary.
External systems become integrations with the system — not hidden dependencies inside individual agents.
This creates a cleaner separation:
Domain
Domain
What the business means.
Kernel
Kernel
How the domain is securely executed.
Agents
Agents
How intelligent decisions and transformations are performed.
Integrations
Integrations
How the domain interacts with the outside world.
Agents can collaborate without becoming the architecture
Agents can collaborate without becoming the architecture
Multiple specialized agents can operate inside the same execution environment.
For example -->
Each agent can specialize.
One can research.
Another can reason.
Another can generate.
Another can validate.
Another can communicate.
But they all operate against the same underlying domain model and execution rules.
Many agents. One system.
Many agents. One system.
AI can be autonomous without becoming uncontrolled
AI can be autonomous without becoming uncontrolled
The objective is not to prevent agents from acting.
It is to give them a well-defined environment in which they can act safely.
An agent may propose an operation, but ...
HexaKernel can determine:
whether that operation exists;
whether the current state allows it;
whether the actor has permission;
whether required conditions are satisfied;
which policies apply;
which side effects should occur;
which events should be generated;
and what must be recorded.
This creates an important separation:
Agents provide intelligence.
The Kernel provides control.
Or even more simply ...
Let agents think. Let the Kernel enforce.
Let agents think. Let the Kernel enforce.
The domain survives the AI stack
The domain survives the AI stack
AI technology is changing at extraordinary speed.
Today's models will not necessarily be tomorrow's models.
Today's agent frameworks will be replaced.
Today's orchestration platforms will evolve.
But the underlying business domain can remain.
A legal case is still a legal case.
A customer is still a customer.
A policy is still a policy.
A transaction is still a transaction.
A business rule remains a business rule.
By placing the domain above the technology stack, HexaKernel allows organizations to evolve their AI infrastructure without continuously rewriting their business systems.
From AI dependency to AI portability
From AI dependency to AI portability
Traditional architectures tend to create dependencies:
The technology serves the domain but the domain does not become a hostage to the technology.
A new economics of software evolution
A new economics of software evolution
This architecture has an economic consequence.
If the domain specification remains stable while the execution technologies beneath it can change independently, organizations no longer need to repeatedly rebuild their applications every time their technology stack evolves.
You can:
Change an agent without changing the domain.
Change a workflow engine without changing the application.
Change an AI provider without rewriting business logic.
Change an integration without redesigning the system.
Adopt new AI capabilities without rebuilding the execution foundation.
And when the domain itself changes, the change can begin where it belongs:
At the specification level.
At the specification level.
The architecture for an AI-native system
The architecture for an AI-native system
HexaKernel is not an agent.
It is not a workflow builder.
It is not an LLM.
It is not another orchestration platform.
It is the execution foundation beneath them.
The upper layer defines what the business is.
The Kernel defines how that domain is securely executed.
The layers underneath provide intelligence, automation and connectivity.
And because they are underneath the Kernel, they can evolve independently.
Build on a stable foundation. Let everything else evolve.
Build on a stable foundation. Let everything else evolve.
The future will not be defined by a single AI model, agent framework or automation platform.
It will be defined by systems capable of continuously adopting whatever technology becomes available next.
HexaKernel is designed around that principle.
**Your domain should be durable.
**Your domain should be durable.
Your agents should be replaceable.
Your technology stack should be evolvable.**
Agents change.
Models change.
Platforms change.
The domain remains.
Put the agents under the Kernel.
Put the agents under the Kernel.
Build software from the logic of your business.
HexaKernel is an execution architecture for domain-driven, AI-native software.
Formalize your domain. Let AI transform it. Let HexaKernel execute it.
ATAVISTICO OÜ
ATAVISTICO OÜ
HexaKernel is a technology product and brand operated by Atavistico OÜ.
© 2026 Atavistico OÜ. All rights reserved.
HexaKernel | ATAVISTICO SPECFORGE
Page updated
Google Sites
Report abuse