01
Persistent Core Memory
Tiered memory retains experience, previous conclusions and corrections, then governs when old work may be reused, expanded, transformed or reconsidered.
AL.I.C.E. · Governed AI architecture
Intelligence
with continuity.
A governed AI architecture designed to remember, retrieve, reason and evolve without losing sight of evidence, trust or human intent.
01 / Memory
02 / Context
03 / Governance
01
The project
Beyond a single prompt.
Beyond a single model.
AL.I.C.E. surrounds an incumbent language model with the systems needed for dependable, persistent work.
Most AI interactions begin almost from zero. AL.I.C.E. is being built around continuity: a durable record of experience, governed access to source material, and a clear path from evidence to answer.
The model remains replaceable. The memory, context, controls and provenance belong to the AL.I.C.E. instance—allowing capability to improve without discarding its history.
02
System architecture
A governed cognitive loop
Context becomes useful
when every step is accountable.
01 / 06
Arbiter
The decision boundary for confidence, trust, grounding and escalation. The Arbiter is designed to prefer supported conclusions—and to say “I don’t know” when the evidence is insufficient.
- Confidence and trust gating
- Provenance-aware decisions
- Controlled escalation
Select a node to explore the system.
Explore the AI architecture
Arbiter
The decision boundary for confidence, trust, grounding and escalation. The Arbiter is designed to prefer supported conclusions—and to say “I don’t know” when the evidence is insufficient.
Discovery Catalogue
A metadata-first view of local and network resources. AL.I.C.E. separates knowing a source exists from locating it and from possessing its contents.
Context Cloud + Forest
The governed home for source-derived knowledge. Sources become structured trees; the forest organises those trees by subject and project for grounded retrieval.
Core Memory
A tiered historical and cognitive feedback loop. It retains experience, conclusions and corrections while trust, freshness and dispute rules govern reuse.
CSP Security
A prompt and content security layer that inspects incoming material before it can influence the system, with normal, override and human-review handling.
Agents + Oracles
A path to specialist and higher-capability models without surrendering local history. Escalated work returns to the same governed context and memory system.
03
Capabilities
What the foundation enables
Designed to know more
than the next likely word.
02
Grounded context
Context Cloud and Context Forest organise source-derived information into navigable trees, supporting multi-document retrieval, citations and evidence-aware answers.
03
Secure ingestion
Material moves through discovery, acquisition, quarantine, validation, transformation and prompt-content security before it can enter the governed context store.
04
Traceable reasoning
An append-only exchange sequence records model and instance identity, timestamps, memory references and source receipts—the basis for answering “How do you know?”
05
Efficient reuse
Cache and memory policies avoid unnecessary inference when trusted work can be reused safely, reducing repeated computation while retaining freshness and dispute checks.
06
Model + agent orchestration
AL.I.C.E. can keep local continuity while routing selected work to specialist agents or more capable “Oracle” models according to cost, trust and capability policies.
04
How it works
From source
to supported answer.
A controlled path preserves identity, evidence and policy from the moment information is discovered to the moment it influences a response.
- 01
Discover
Catalogue available resources and changes without assuming that discovery grants permission to ingest.
- 02
Protect
Acquire on demand, quarantine content, validate identity and apply CSP prompt-content controls.
- 03
Contextualise
Transform admitted material into source trees and governed context packages with revision-bound receipts.
- 04
Reason
Combine relevant evidence with useful prior experience while keeping source data and remembered conclusions distinct.
- 05
Govern
Apply trust, sufficiency and escalation rules before a conclusion is used or an action is considered.
- 06
Remember
Record what happened, what was concluded and why—creating continuity for the next interaction.
The long view
“The intelligence is not only in the answer. It is in the continuity between answers.”
AL.I.C.E. treats memory, evidence and governance as first-class parts of cognition—not accessories added after generation.
05
Project direction
Foundation in active development
Built in layers.
Proven as a system.
The Year One foundation establishes the major architectural blocks. Current work is focused on making their relationships increasingly authoritative, observable and production-ready.
Established foundation
- Core Memory and continuity
- Discovery and governed ingestion
- Context Cloud and grounded retrieval
- Security and provenance foundations
- Tree Connectome and visualisation
Advancing next
- End-to-end subsystem integration
- Arbiter and Dreaming modernisation
- Forest-scale routing and relationships
- Operational tooling and observability
- Governed distributed agent execution
AL.I.C.E. is an active research and engineering project. Capabilities described here include established foundations and continuing development, not a claim of production completion.
06
About & contact
The home of the AL.I.C.E. project
About Aeonworx
Aeonworx is the home of AL.I.C.E., an ongoing AI research and engineering project exploring persistent memory, grounded context and accountable reasoning.
The project's foundation brings together memory, source discovery, governed ingestion and provenance. Development continues towards integrating these systems into a coherent architecture that can retain context and explain where its answers came from.
Contact Jeff
For enquiries about this website or the AL.I.C.E. project, contact Jeff at jeff@aeonworx.com.