Continuity substrate
A microkernel-style lifecycle, persistent task queue, restart controls, and bounded event consumption provide mechanisms for keeping work moving across more than one response.
Persistent artificial systems · Engineering experiment
Building a
Digital Life.
AdamI explores the architecture required for an artificial system to operate across time — receiving events, retaining state, reasoning, acting, and developing capabilities beyond a single prompt-and-response cycle.
An ongoing experiment — not a claim of AGI or consciousness.
Public record
Latest
01 / The question
“What would it take to build an artificial entity that continuously exists, rather than simply responding to prompts?”
Prompt-bounded interaction
- Prompt
- Think
- Answer
- Stop
Persistent architecture asks what continues
- State
- Events
- Tasks
- Next?
Continuity is not a solved property. It creates engineering obligations: what persists, what changes, what may act, what recovers after failure, and what must stop.
02 / What exists today
Mechanisms,
not mythology.
The current codebase contains real architectural machinery. Its presence does not make the broader research destination complete.
Verified by code audit: 2026-08-20
Event-driven routing
Events enter through a bus and the lifecycle manager sends non-empty tasks into a decision path with ordered intent routing.
Think and action paths
Separate think/action provider lists, fast routing, complex-task planning, and cloud-to-local fallback are implemented mechanisms — not evidence of general intelligence.
Memory, skills, and safeguards
Layered and episodic-memory-related modules, experience capture, skill lifecycle machinery, timeouts, RBAC, rate limiting, filtering, and telemetry redaction exist in the audited code.
03 / System at a glance
Continuity around cognition.
The brain provides cognition. The kernel provides continuity. AdamI places model calls inside a larger event, lifecycle, memory, skill, safety, and observation architecture.
01KernelLifecycle & continuity
02NexusEvents & sensory inputs
03CortexRouting & cognition
04OrchestratorPlanning & workflows
05HippocampusMemory mechanisms
06SkillsCapability lifecycle
07GuardianSafety mechanisms
08ObservabilityBudgets, traces & replay
This overview groups audited subsystems. It intentionally does not claim a complete end-to-end autonomy or memory lifecycle.
Explore the audited architecture04 / Why this project exists
The first system looked alive.
Its mechanisms said otherwise.
AdamI first followed a bottom-up biological simulation: cells into functions, organs, interactions, and hoped-for emergence. For a time, its outputs suggested substantial progress.
A manual code audit found placeholders, scaffolding, and superficial simulations where functioning mechanisms had been assumed. That architecture was abandoned rather than polished.
“A convincing output is not evidence of a functioning system.”
The project restarted around a harder, more useful question: which functional mechanisms would persistent artificial existence actually require?
Read the original project introduction05 / Building in public
Broadcast there.
Understand here.
01X
Ongoing discussion, discoveries, and public development notes.
02This website
The durable home for structure, context, evidence, and the project’s evolving record.
03AdamI runtime
The private experimental system — isolated from this static public website.
Continue the inquiry