Deterministic = Code
If it must happen, enforce in code. If it requires interpretation, use AI. Safety gates and routing rules are never prompted.
Architecture
Five layers, from values to execution.
Coordination technology optimizes execution, not direction. Exuno starts with what matters and treats execution as a consequence.
L5
Values & Preferences
Purpose, standards, tradeoffs, and judgment calls gathered through onboarding interviews and approved owner feedback.
L4
GBrain Memory
Approved source pages, entity records, concept maps, and searchable memory with source boundaries preserved.
L3
Hermes Agent Routing
Routing judgment for which agent or local tool should receive a task, with recommendations queued before action.
L2
Claude Code and Codex Selection
Choosing the right Claude Code or Codex path, breaking work into plans, and assigning focused agents.
L1
Claude Code, Codex, and Subagents
The local execution layer for coding, extraction, search, summarization, checks, writing, and analysis.
Principles
What survived twenty iterations.
Two-Way Information
Every agent returns work and a reality report — confidence, blockers, assumptions. Problems are assets to surface.
Should vs. Can
“Should we?” is evaluative — humans decide. “Can we?” is epistemic — AI discovers. Confusing these causes drift.
Psychological Safety
“I'm not sure” beats a plausible guess. Fabrication is the cardinal error. Systems that surface uncertainty are trustworthy.
Lazy Managers
Managers delegate and orchestrate, never execute directly. The constraint forces clarity of intent before work begins.
Untouchable Values
Your values sit above the system. Professional judgment remains sovereign — AI-optimized within boundaries, not moved by them.
Implementation
A practical framework.
Map Your Craft
What requires judgment vs. what is mechanical? Where does institutional knowledge live? The audit reveals where a twin adds value.
Build the Knowledge Base
Start with one person, one domain. Ingest existing work into structured records. Goal: a single source of truth any agent can read.
Connect the Agentic Layer
Wire up automated intake. Build the first workflow. Let the system learn from corrections. This is where compounding starts.
Why
The architecture is a philosophical argument.
AI externalizes cognition — which is the opposite of maintaining awareness. The question is whether we extend our judgment or replace it. Exuno's answer is structural: values sit above the system because they can't be codified into it.
The Third Enlightenment
The thesis. Powerful tools require maintained awareness, not less of it. Acting with full commitment while holding premises lightly.
What Rules Can't Capture
Why L5 sits above the system. Tacit knowledge resists codification — you can't prompt your way to judgment.
The Right Direction
Why agents need a sovereign layer. Coordination optimizes execution, but someone still has to choose which game to play.
Ready to build your twin?
Download Exuno for a local-first twin scaffold, or work with us to build it for your team.