Intent
Guardian starts from what a person is trying to achieve — the outcome, the constraints, the people affected — rather than an isolated command.
How Guardian works
Intelligence with a boundary around it.
Guardian is an operating architecture for human life: it understands intent, remembers with permission, reasons carefully, and acts only inside limits a person set.
The lifecycle
Intent → Memory → Intelligence → Permission → Action → Verification → Learning.
Each stage exists so the next one can be trusted. The chain is intentionally interruptible by a human at any point.
Memory
Relevant context is recalled from memory the person has permitted, scoped to the situation at hand and visible on request.
Intelligence
Guardian reasons about options, timing, dependencies and trade-offs, and states what it does not know.
Permission
Before anything leaves the system, Guardian checks whether it holds the authority to act, and at what scope.
Action
Permitted actions are carried out narrowly and recorded, with the reason and boundary attached.
Verification
Guardian confirms what actually happened, reports honestly when it did not, and never assumes success.
Learning
Corrections improve future judgment. Learning never silently widens what Guardian is allowed to do.
Memory Fabric
Context that belongs to the person, not the product.
Memory Fabric is Guardian's permissioned understanding of a life or an organization: commitments, relationships, preferences, documents and history.
- Every memory has a visible origin and a purpose.
- Scopes stay separate: personal, family, household, work.
- People can inspect, correct, limit, delete and export.
- Sensitive categories are handled with heightened restraint.
Intelligence Fabric
Reasoning that shows its footing.
Intelligence Fabric turns context into judgment: what matters now, what depends on what, what is uncertain, and what a reasonable next step looks like.
Relevance
Surfacing the few things that change outcomes today.
Dependency
Understanding what is blocked, and by whom.
Uncertainty
Saying plainly when confidence is low.
Intent specification
Say the outcome; Guardian holds the shape of it.
An intent captures the goal, the constraints, the people involved and the limits of authority — so guidance stays coherent as details change.
Durable
Intents persist across days and tools instead of dying in one thread.
Editable
Change the intent and everything downstream re-aligns with it.
Permission & action boundaries
Knowing something is not permission to do something.
Guardian holds two separate authorities. Knowledge authority governs what it may understand. Action authority governs what it may do — and it is always narrower.
Observe
Understand only; take no action.
Suggest
Recommend, with reasoning attached.
Prepare
Draft and stage, pending human approval.
Act (scoped)
Execute a specific, granted, revocable action.
Human review & correction
Correction is a first-class interaction.
When Guardian is wrong, saying so should be easy, immediate and consequential. Corrections change the record and the future behaviour.
- Any surfaced item can be questioned, edited or dismissed.
- Corrections are stored with the memory they correct.
- Repeated corrections adjust judgment, not authority.
- Actions carry a record a human can review after the fact.
Continuous learning
Better judgment. Never wider authority.
Guardian improves at understanding you over time. What it is allowed to do changes only when a human explicitly changes it.
This page describes Guardian at a conceptual level. Implementation specifics, internal policy mechanics and proprietary methods are intentionally not published here.