Multi-agent Environment of EduOS Lad
Specialist AI agents support education, research, project work and the management of university processes.
EduOS distributes navigation, tutoring, analysis, critical review, coordination, resource and competence support, and governance among specialised functions. Each agent receives bounded context, defined expectations and rules. People remain accountable for decisions.
A developing research architecture, illustrated by an interactive model, not an institution-wide operational deployment.
Navigator
Finds available actions and the next step.
- 01Event → context → task
- 02Select agents and route
- 03Review evidence and proposed changes
- 04Accountable decision → action → result
Specialisation and traceable outputs
- 01
One defined function
Each agent performs a recognised class of actions.
- 02
Bounded context
Agents receive the data necessary for a particular task.
- 03
Defined output
Every output has a known format and recipient.
- 04
Observable history
Inputs, actions, outputs and transitions can be reviewed.
- 05
Governed rules
Criteria and permissions are set by the process owner.
Five levels of the agent environment
Incoming task context is routed to specialist agents, checked against criteria and passed to an accountable participant.
Incoming context
Structure the task, role, data and constraints.
Eight specialised agent functions
Each agent has bounded inputs and outputs, a recipient and a responsible human reviewer.
Navigator
Finds available actions and the next step.
Routing a task between agents
Route choice depends on task type, permissions, goals, risk, data and resource constraints. A parallel route means independent checks before coordination, not a sequential handoff.
Simple
- 01Task
- 02One agent
- 03User
Example: Find useful material for a learning topic.
Agents help a team overcome a project delay
A student team is behind schedule and its interim outputs conflict. Agents analyse the problem and prepare options for the team and authorised decision-makers.
01 · Detect signal
Context and memory boundaries
An agent receives information necessary for its function under the user's permissions. Historical memory does not automatically confer broader rights.
- 01
User
- 02
Task
- 03
Subject
- 04
Process
- 05
Governance
Memory levels
- 01
Session context
- 02
Process history
- 03
Role profile
- 04
Institutional knowledge
Information permitted to each agent
Participant roles and authorised help
Student
Reviewing the quality of an agent result
Example: an agent proposes changing roles in a project team. Each criterion identifies a distinct question, evidence and reviewer.
Correct function
Output addresses the agent's assigned function.
Governing an agent throughout its lifecycle
Define function
An energy-saving project with multiple agents
A demonstration project develops a model to reduce energy use in a university building; people review the model and authorise a pilot.
01 · Task arrives
University operations requests a model for reducing building energy use.
What makes agent work accountable
- 01
Specialisation
Each agent has an explicit function and type of output.
- 02
Routing
EduOS connects specialist agents according to the task's context.
- 03
Bounded context
Access follows the information and permissions needed for a function.
- 04
Review
Recommendations pass criteria, critique and accountable approval.
- 05
Provenance
Inputs, agent actions, recommendations and decisions remain connected.
- 06
Human accountability
Educational, expert and governance decisions belong to authorised roles.
- 07
Governed lifecycle
Agents are designed, tested, introduced, observed and revised under explicit rules.
About the model
This is a research architecture developed by Sergey Shatokhin as part of University 5.0 at IWXA. Agent specialisation, task routing, limited context, quality review and human responsibility are connected in one proposed environment.