Governance is the architecture.

Current verified behavior

IEVA’s working prototype is a single-user, local-first knowledge system. The following behaviors are verifiable in that system today:

• Human-readable Markdown is the source of truth — all memory and context can be opened and read directly.
• Version control (Git) records the history of changes to memory, skills, and content, providing provenance and reviewable diffs.
• Source links connect derived knowledge to where it came from.
• Workflows are explicit, version-controlled skill documents rather than opaque behavior.
• Significant decisions and irreversible actions require human approval in the operating routine.
• Evergreen knowledge is harvested only from completed, shipped work — aspirational notes are not promoted to fact.

Product design principles

These principles govern the commercial product’s design. They are commitments about how IEVA is being built, stated as principles — not as descriptions of a shipped enterprise platform.

Least-privilege access

IEVA reads only sources you explicitly authorize, with the narrowest scope that serves the task.

Explicit source authorization

Nothing is connected by default. Every source is a deliberate, visible, revocable decision.

Human approval gates

Consequential actions — sending, committing, spending, deciding — wait for a person. No code path routes around the gate.

Observable actions

Tool use is visible: what ran, when, against which sources, producing what.

Source provenance

Derived context, briefing items, and delegated results keep citations to their origins. The source of a recommendation never disappears.

Reviewable memory

Memory is presented for inspection — organized, attributed, and readable — never as an opaque store.

Memory correction and deletion

Wrong or stale beliefs can be corrected or deleted, and corrections propagate with their own history.

Bounded agent delegation

Delegated work carries a clear objective, access limited to the task, and a reviewable handoff.

Execution isolation

Delegated tasks run in isolated working contexts so work cannot contaminate memory or other tasks as a side effect.

Auditability

Actions, approvals, and memory changes accumulate into reviewable history as a side effect of normal operation.

Retention and data control

Memory and derived context follow explicit retention rules, and disconnecting a source is a first-class operation.

Reversibility where possible

Operations are designed to be undoable where the underlying action permits; irreversible actions earn stricter gates.

Planned and conceptual controls

The following are planned for the product platform and are not yet implemented or verified: role-based permissions for teams, policy-aware action controls, organization-level audit trails, isolated execution environments at platform scale, and data residency and retention tooling. We will move items from this list to “verified” only when they exist and have been tested — not before.

We make no claims of SOC 2, ISO 27001, GDPR certification, specific encryption implementations, or data-residency guarantees at this stage. If a security review requires specifics, ask us directly — you will get honest answers about maturity.

Security contact

To report a vulnerability, ask a security question, or request detail for a review, contact hola@aekora.com with “Security” in the subject line. A dedicated security address will be published when available.