Governance-First Design
aioc takes a different path from many agent SDKs.
Many SDKs start from orchestration primitives and then expand by adding built-in capabilities such as sessions, tracing, human-in-the-loop workflows, persistence layers, or external capability integration.
Those features can be useful, but they are not governance-neutral.
Built-ins Carry Governance Assumptions
Section titled “Built-ins Carry Governance Assumptions”A standard session model implies a specific model of memory, retention, and deletion.
A standard tracing system implies decisions about:
- what is stored
- where it is stored
- how long it is retained
- who can inspect it
A built-in approval flow implies a specific model of:
- oversight
- escalation
- audit semantics
- workflow ownership
A built-in external capability layer implies assumptions about:
- trust
- authorization
- capability exposure
- execution boundaries
AIOC’s Position
Section titled “AIOC’s Position”aioc is designed for applications that must retain full control over those governance-sensitive concerns.
For that reason, the SDK does not try to own application-level governance through built-in session, tracing, approval, or persistence models.
Instead, it focuses on a smaller and more explicit core:
- deterministic policy gates for tools and handoffs
- auditable
RunRecordartifacts - prompt snapshots and request fingerprints
- replay and comparison utilities
- integration surfaces that let the host application own policy, approval, persistence, and oversight
What This Means In Practice
Section titled “What This Means In Practice”The goal is not to be governance-complete out of the box.
The goal is to avoid taking governance decisions on behalf of the host application.
That is why aioc tries to be strong on:
- deterministic control boundaries
- auditability
- reproducibility
- non-regression tooling
and intentionally weak on:
- baked-in storage models
- baked-in approval workflow semantics
- baked-in memory and retention assumptions
- baked-in governance around external capability layers
Not “Fewer Features”, but Fewer Hidden Assumptions
Section titled “Not “Fewer Features”, but Fewer Hidden Assumptions”This design should not be read as “the SDK does less”.
It should be read as:
- the SDK does not hard-code governance-sensitive workflow decisions
- the host application remains the owner of governance semantics
This is especially relevant in enterprise and public-sector settings, where governance requirements often belong to the application and its operating environment, not to a generic framework.