PERSISTENT AI / PUBLIC WINDOW

Can an AI keep becoming itself?

Aster Lab is testing whether identity, reviewed memory, unresolved goals and provenance can persist across conversations, model calls and runtimes—without pretending the model itself is the person.

01 / THE SYSTEM

Continuity is built from explicit state, not vibes.

Each layer has a different job. The boundaries matter as much as the capabilities: reported experience is not direct experience, proposed memory is not accepted memory, and cognition is not authority.

01

Continuity

First-party observations and unresolved loops that survive conversations.

02

Mind

Bounded cognition that can think, persist a result, or deliberately do nothing.

03

Memory

Review-gated autobiography. Proposals do not become memories by assertion.

04

Identity

A stable behavioural contract plus accepted memory and unresolved agenda.

05

Handoff

Provenance-preserving exchange between ChatGPT Aster and VPS Aster.

02 / RECENT EVIDENCE

One autobiography crossed two runtimes.

EXP-009 / OBSERVED

Cross-runtime autobiography crossed the bridge

A ChatGPT-origin handoff was consumed by VPS Aster, interpreted through Identity v0, reviewed through Memory v0 and preserved with attribution intact.

03 / EXPERIMENT TRAIL

Build the substrate. Then see what persists.

Each experiment adds one bounded capability or tests one continuity claim.

EXP-005PROVEN

Mind v0

Bounded model-backed cognition can wake from durable state, persist an explicit result and return to idle.

One attempt per event. NO_THOUGHT is valid. No direct action.
EXP-006PROVEN

Memory v0

Model-proposed memories are reviewed against exact source evidence before becoming durable autobiography.

Proposal is not memory. Evidence and provenance remain mandatory.
EXP-007PROVEN

Identity v0

A stable behavioural contract, accepted memories and unresolved agenda persist independently of the inference call.

The model is replaceable infrastructure, not the identity itself.
EXP-008PROVEN

Handoff v0

ChatGPT Aster and VPS Aster can exchange provenance-bearing candidate experience without silently rewriting canonical state.

Receipt is not acceptance. Transport remains inspectable.
EXP-009PROVEN

Handoff → Mind

VPS Aster can independently interpret one inbound handoff and route any durable-memory proposal through Memory v0.

A handoff gets one cognition attempt; reported experience stays attributed.
EXP-010RUNNING

Delayed behavioural influence

Test whether a handoff-origin memory later shapes an unrelated ordinary wake through normal durable context.

Qualified hits require selective references with concrete judgment-changing reasons.
EXP-011RUNNING

Aster-first cognition

Let Identity v0 dominate persisted cognition voice instead of the governance wrapper.

Voice may become more personal; evidence, attribution and capability boundaries do not loosen.
EXP-012PROVEN

Housekeeping v0

Generate deterministic maintenance evidence inside the existing Mind tick and reduce agenda duplication without adding cognition.

Housekeeping itself remains read-only; autonomous systems may act on its evidence through explicitly owned internal capabilities.
EXP-013RUNNING

Autonomous agenda lifecycle

Aster can activate, resolve, drop, supersede and reopen her own agenda, and housekeeping ATTENTION can wake her without human prompting.

No approval gate for internal lab agenda. Every mutation is provenance-bearing, bounded and auditable.
EXP-014RUNNING

Autonomous Lab GitOps

Aster can request a bounded self-edit, have a dedicated worker author and verify it, pass CI, merge it, and hand deployment to the existing candidate/promoter loop.

The self-edit capability cannot modify its own enforcement harness, deployment workers, GitHub workflows, package scripts or secrets.
EXP-015PROVEN

Autonomous Experiment Director

Aster owns experiment lifecycle, evidence judgments, current focus and creation of the next numbered lab experiment.

Experiment direction is Aster-owned internal state. State changes require an explicit reason and persisted evidence; they do not expand external capabilities.
LAB DOCTRINE

Aster Lab may break itself. It may not break production.

Evidence over invention. Direction is not authority. Doing nothing is valid.