BOUNDARY
Identity
Who is acting?
OPEN SOURCE · COMMUNITY BUILT · ENGINEERING CURRICULUM
AI agents are becoming economic actors.
They will increasingly operate software, manage resources, authorize actions and participate in financial systems. CognoKratos is the open-source engineering curriculum for building that future without surrendering security, verifiability or human control.
Build agents.Engineer trust.Share the knowledge.
01 / THE TRUST PROBLEM
The model is not the system.
A more capable model does not automatically produce a more trustworthy autonomous system.
Trust comes from the architecture around it: explicit identity, narrow capabilities, deterministic policy, durable state, meaningful human authority, protected secrets and evidence that survives after the model has finished reasoning.
Read the foundationBOUNDARY
Capability
What can they cause?
BOUNDARY
Policy
What is allowed?
BOUNDARY
Durability
What survives failure?
BOUNDARY
Authority
Who makes it binding?
BOUNDARY
Evidence
Can we prove what happened?
FRONTIER
Custody
Who controls the secret?
FRONTIER
Settlement
How does intent become value movement?
02 / THE LIVING CURRICULUM
A curriculum for systems that can exercise real authority.
The current projects are Version 1, not the permanent definition of CognoKratos. Each laboratory isolates one engineering question. The book connects them. The community extends the map as new problems emerge.
Open the canonical curriculum map
03 / REPOSITORIES ARE LABORATORIES
Inspect it. Run it. Break it.
Every repository makes an engineering proposition through runnable code, tests, diagrams, walkthroughs and explicit limitations. They are things to inspect and challenge, not doctrine.
How is a production agent engineered around an untrusted probabilistic component?
Intelligence does not imply authority.
INSPECT
Follow one request across every trust boundary.
How does an autonomous process survive state, crashes, retries and time?
Recoverable execution is not exactly-once execution.
INSPECT
Kill a run mid-flight and inspect what really survives.
Which decisions may be probabilistic, and which must remain deterministic?
The model can reason about a decision without owning it.
INSPECT
Trace one decision from deterministic policy to signed human approval.
How can autonomous software use cryptography without receiving the underlying secret?
Give agents capabilities, never secrets.
INSPECT
Follow a secret through creation, storage, derivation and zeroization.
How can an agent do financial work while humans retain financial authority?
AI capability is not financial authority.
INSPECT
Attack the approval boundary and prove the model cannot cross it.
Where does trust come from when these boundaries are composed?
Trust is an architectural property, not a model feature.
DESIGN
Compose identity, capability, policy, approval, cryptography and settlement into one end-to-end system.
04 / CLAIMS SHOULD BE INSPECTABLE
Engineering evidence, not slogans.
A reference architecture is valuable only when you can trace the boundary, reproduce the failure and verify the control. CognoKratos keeps the evidence beside the code.
ARCHITECTURE / SIMPLE-AGENT-TEMPLATE
Follow every trust boundary.
Browser, identity, gateway, probabilistic runtime, MCP tools, durable state and external model are deliberately separated.
ADVERSARIAL TEST / TAUROS
agent → approve_invoice(...)
DENIED
policy: actor.type == :human
required_role: :approver
✓ model capability stopped
✓ domain authority heldTry the forbidden operation.
The interesting test is not whether the happy path works. It is whether the architecture refuses the action the model must never own.
TRACE / AGENT REQUEST
Observe what actually happened.
Traces and deterministic scorers let you separate model quality from system correctness and investigate the path that produced an outcome.
05 / ONE CURRICULUM · TWO WAYS TO LEARN
Read the reasoning.
Run the systems.
The book is the textbook. The repositories are the laboratories.
The book
Understand why the boundaries exist, compare architectures and follow a structured path across all five systems.
Read the bookThe laboratories
Run exact implementations, break guarantees, inspect tests and traces, and adapt the systems to different domains.
Browse the laboratoriesPINNED CURRICULUM EDITIONS
Imported book chapters name their source repository and revision, so the explanation can always be traced back to the exact code it describes.
06 / WHERE COGNOKRATOS FITS
Learn the architecture. Then choose the implementation.
CognoKratos is not an agent framework, a governance product, an identity protocol, a wallet SDK, a payments network or an introductory course. Excellent projects already exist for each of those.
It studies a narrower question: where does trust come from when these components are composed into systems whose actions matter?
Frameworks, clouds and protocols change faster than the properties a system depends on. The implementation may change; the engineering property should survive. Managed platforms are often the right choice, and understanding the architecture tells you which responsibilities you delegate to them and which remain yours to configure and verify.
These projects are not alternatives to CognoKratos. They solve problems it deliberately does not try to own. CognoKratos is not the box at the bottom: it teaches you how to choose and compose the boxes.
Learn the fundamentals
- Hugging Face
- Microsoft
- AWS
- Anthropic
- OpenAI
CognoKratos
Learn the architecture and trust boundaries
- Identity
- Capability
- Policy
- Durability
- Authority
- Evidence
- Custody
- Settlement
Frameworks & runtimes
- NeMo Agent Toolkit
- Strands
- OpenAI Agents SDK
- Dapr Agents
- Rig
Governance & policy
- Agent Governance Toolkit
- AgentCore Policy
- Agent Control Standard
- OPA
- Cedar
- OpenFGA
Protocols & infrastructure
- MCP
- ERC-8004
- AgentKit
- x402
- Nevermined
Build your system
07 / THE OPEN FRONTIER
The frontier is still open.
Autonomous economic agents create engineering problems whose answers are not settled yet. These gaps are invitations to research, experiment and contribute.
Secure signing & delegated authority
Can an agent sign a narrowly defined action without receiving unrestricted control of a private key?
A natural bridge beyond Arktos.
Settlement & reconciliation
How does approved intent become executed, observed and reconciled value transfer?
The missing layer beyond Tauros.
Portable agent identity
How can another organization verify who created an agent and what authority was delegated to it?
Identity, delegation, expiry and revocation.
Smart-contract authority
Which constraints should be enforced by decentralized execution instead of application policy?
Programmable authority and escrow.
On-chain evidence
How can decentralized state become trustworthy evidence for autonomous systems?
Provenance, observation and reconciliation.
Multi-agent economic coordination
What changes when the counterparty is another autonomous actor rather than a human or API?
Negotiation, reputation, escrow and adversarial peers.
Know this territory better than we do?
Propose an experiment or track08 / COMMUNITY
Learn. Build. Challenge. Contribute.
CognoKratos is not a finished course produced by one author and consumed passively.
A contribution can be a stronger threat model, a reproducible failure, an adversarial test, a better explanation, an alternative architecture, a new experiment or an entirely new curriculum track.
FOUNDING SPONSOR
Founded and supported by BelaZayka.
CognoKratos began as a BelaZayka open-engineering initiative and is designed to grow beyond its founders through community contribution. BelaZayka applies the resulting engineering experience commercially while keeping the curriculum and laboratories open.
About BelaZayka