CTO / Tech / BA Onboarding

Aevum is a local-first personal intelligence system, not a generic AI app.

The team must build from ground truth: one canonical ingestion model, honest system behavior, memory continuity, and a companion experience that becomes more valuable as a person keeps speaking.

Success is not “features shipped.” Success is a system where capture, memory, retrieval, and response feel coherent, trustworthy, and compounding.
System Boundaries

What Aevum must be, and what it must not drift into.

Must be

  • Local-first
  • Deterministic where possible
  • Truthful in UI and system behavior
  • Centered on memory continuity
  • Apple-native in execution model

Must not become

  • A fake-cloud hybrid with misleading UX
  • A fragmented bundle of unconnected capture paths
  • A shallow chatbot that ignores the graph
  • A UI with visible actions that have no real backend behavior
Delivery Priorities

The implementation priority is coherence, not breadth.

1

Canonical ingestion

All first-party inputs must land through one approved contract and create visible root entries immediately.

2

Truthful graph behavior

Enrichment, linking, and scheduled processing must be real or explicitly absent. No implied automation without implementation.

3

Grounded intelligence

Responses must use persona, session, and local memory context. No dummy state. No shallow filler as default.

If a button exists, it must work. If a process is described, it must be implemented. If a capability is not real, the UI must not imply that it is.
Guardrails

The team must prevent standard LLM implementation mistakes.

Absolute guardrails

  • No duplicate write paths
  • No orphan UI actions
  • No placeholder logic in production flow
  • No inconsistent provenance mapping
  • No cloud/off-device implication unless approved

Common failure modes to avoid

  • Assuming architecture instead of tracing real code
  • Calling a queue “done” without visible user result
  • Adding enum cases without full routing
  • Writing tests that validate helpers instead of live orchestration
  • Letting companion responses degrade into generic therapy-speak
Team Contract

How Antigravity, CTO, engineers, and BAs should work together.

Build in chunks

  • Compile-safe foundation first
  • Then routing
  • Then visibility
  • Then intelligence depth
  • Then validation

Return evidence, not confidence

  • Exact files changed
  • Exact paths rewired
  • Exact tests added
  • Exact proof traces
  • Exact known limits remaining
The operating principle is simple: reality before narrative. Establish the real system, then improve it. Never invert that order.
Theme based on codebase tokens: #0A84FF Aevum Blue, #0B0B0F primary background, #222222 dark neutral.