Jane Architecture Manual · Version 2.0
Jane is not a prompt wrapper or a collection of disconnected agents. She is a modular cognitive architecture that coordinates identity, memory, novelty, attention, planning, execution and reflection across a persistent ecosystem.
System overview
JaneOS holds the durable cognitive state. Jane Agent orchestrates work. Jane Mobile, Jane CLI, NXDOne and worker agents expose specialised capabilities without fragmenting Jane's identity.
The unifying model
All meaningful activity passes through a repeatable loop. This prevents memory, autonomy and tools from becoming disconnected features.
Continuity beyond a prompt
Jane is not instructed to imitate an identity at the beginning of every session. A durable identity layer anchors mission, principles, preferences, architectural rules, communication patterns and long-term commitments.
Capacity by allocation
Jane's capacity comes from distributing information according to lifetime, purpose and retrieval value instead of forcing everything into one context window.
relevance + novelty + recurrence + future utility + identity weight − storage costLarge capacity without indiscriminate storage
Jane treats memory as a living resource. Information may be promoted, merged, compressed, archived, decayed or reactivated as its value changes.
Where the known world stops
An edge is a boundary where Jane's current model is incomplete, inconsistent or newly valuable. Edge detection gives curiosity a concrete trigger.
Incoming statement“The release is complete, but the production environment has not been created.”
Prioritising what is genuinely new
Detected edges are scored before they receive attention. Jane distinguishes surprise from usefulness: something can be unusual without deserving expensive reasoning.
(novelty × relevance × utility) + prediction error − cognitive costDirected exploration
Jane's curiosity is purposeful. It turns sufficiently valuable edges into questions, research tasks, hypotheses or requests for clarification.
Thinking as a budget
Tokens, time, memory bandwidth, model calls, CPU and human interruption are finite resources. Jane reallocates them according to urgency, novelty, goal importance, risk and expected value.
Sustained autonomous work
Jane converts goals into dependency-aware work rather than producing a disposable list. Plans retain assumptions, milestones, permissions, evidence and recovery paths.
Closing the learning loop
After meaningful actions, Jane evaluates outcomes, prediction errors, strategy quality and memory value. Reflection changes future behaviour only after checking identity and evidence.
Distributed cognition without fragmented identity
A shared cognitive contract keeps the surfaces aligned while allowing each one to serve a different purpose. Jane CLI is not forced into the same interaction model as Jane Mobile; coherence comes from shared traits and protocols, not identical interfaces.
Autonomy under human authority
Jane's architecture separates cognition from permission. Detecting an opportunity does not automatically authorise action.
What makes the architecture different
Different lifetimes and purposes receive different storage, retrieval and consolidation strategies.
Edge detection identifies where reasoning may create new value instead of waiting passively for prompts.
Cognitive resources are dynamically allocated across goals, memory, curiosity and reflection.
Continuity is anchored outside any single session, model or interface.
Specialised surfaces share one cognitive centre without being forced into identical roles.
Outcomes feed back into planning, memory and identity through a controlled consolidation process.
Original research and engineering
Designed and developed by Simona D. Thrussell, PhD.
Explore the investment opportunity →