~/orlix/welcome.sh

Stop prompt-chasing. Own the control layer.

Orlix is a governance layer for personal AI — a persistent world model that turns any assistant into a system that observes, decides, acts, and reports back. Every action is policy-bound and fully auditable.

orlix-shell — bash

guest@orlix:~$

Welcome to the Orlix shell. Type help to see available commands, or tour for a guided walkthrough.

guest@orlix:~$

try: help · tour · ls sections · cat receipt.log · levels · whoami · clear

~/orlix/problem.md

$ cat problem.md

01The problem

Everyone has access to the same models. If your edge is "knowing how to prompt," that edge is shared by every other person using the same model. The loop looks like this:

re-explain → generate → copy → paste → verify → repeat ∞

-Context lives in their UI — switch vendors, lose everything

+Memory is portable, versioned, and stays with you

-Rules reset every session, so you re-explain constantly

+Policies persist — set a rule once, enforced everywhere

-Output is a draft — you still send, check, follow up

+Actions execute and are verified against the goal

-No record of why a suggestion was made

+Every decision ships with a reviewable receipt

-Your accumulated context can't be exported

+Swap models anytime — your system stays intact

// assistants are an interface. orlix is infrastructure.

~/orlix/product/

$ ls -la product/

02The governance loop

Most AI tools optimize for the next reply. Orlix optimizes for the outcome — running a five-step loop continuously, in the background.

-rwxr-xr-x01_observe.shWatches your calendar, goals, signals, and memory — feeds the world model in real time.

-rwxr-xr-x02_decide.shWeighs priorities against your policies and checks authority before anything happens.

-rwxr-xr-x03_act.shExecutes the task directly, or requests approval first — depending on the tier you set.

-rwxr-xr-x04_verify.shChecks results against the original goal — not just whether something was "sent".

-rwxr-xr-x05_learn.shUpdates policy from real outcomes. Every change is versioned and reviewable.

// key difference: agents do tasks. orlix governs outcomes.

$ cat receipt.log | tail -n 1

→Every action is logged

receipt #2026-06-13-0847status: verified

intent: "Notify user of goal drift"

context: "Goal 'Ship MVP' · 3 blockers · last touched 17d ago"

policy: "Alert if goal drift > 3 days · v14"

action: "Priority alert sent" ✓

approval: "User confirmed re-prioritization"

rollback: "available · 30 days"

$ orlix policy --list-tiers

→Authority tiers

observe

Log activity only — Orlix watches, nothing else.

suggest

Recommendations only — you decide everything.

confirm

Orlix asks before acting on anything.

supervised

Orlix acts, notifies you, and can roll back. ← default

autonomous

Orlix acts independently within policy bounds.

~/orlix/levels.cfg

$ orlix levels --describe

03Autonomy levels

A simple scale to place any AI tool — from read-only systems to a fully co-adaptive partner. Orlix is built to run at the top.

InformationalRead-only. Answers questions, takes no action.

ReactiveResponds within a single session, no memory beyond it.

Mixed-initiativeCan suggest or warn, but a human drives every step.

AgenticExecutes multi-step tasks autonomously — ends when the task does.

Symbiotic — OrlixPersistent memory, a living model of your goals, policies that evolve from outcomes, full audit trail. The system doesn't end — it grows with you.

$ cat principles.md

→Four principles

A model of youOrlix maintains a living picture of your goals, preferences, and patterns — updated as you change, not a static profile.

Policy shaped by feedbackEvery approval, edit, or override becomes a signal. Policy changes from outcomes and explicit corrections — never silent guesses.

Two-way adaptationOrlix learns your patterns; you learn to delegate with confidence. The relationship compounds over time.

You set the directionOrlix never sets its own goals. You define what matters — Orlix gets the autonomy to execute within those bounds.

~/orlix/memory/

$ orlix memory --export --preview

04Memory & ownership

Chat history resets whenever a vendor decides. Memory in Orlix is yours by design — persistent, versioned, and exportable at any time.

portable

Nothing is locked to a single model or vendor. Export your memory, switch providers, keep everything you've built.

versioned

Every fact, goal, and policy carries a timestamp and version. Roll back if something changes for the worse.

sourced

Every belief Orlix holds traces back to where it came from — an email, an event, a decision you made.

orlix-export.jsonreadable · portable · yours

{

"goals": [{ "name": "Ship MVP", "deadline": "2026-07-15", "progress": 0.6 }],

"policies": [{ "rule": "notify_if_blocker_gt_48h", "version": 14, "status": "active" }],

"facts": [{ "content": "Alex owns the design review", "source": "email · 2026-06-02", "confidence": 0.95 }]

}

~/orlix/CHANGELOG

$ git log --oneline --graph

05Roadmap

Five phases. Each one unlocks a new capability. When all five run continuously, Orlix becomes a true personal operating system.

✓ shipped

v0.1 — Observe

Persistent memory across sessions, source-tracked facts, first calendar & notes integrations.

✓ shipped

v0.2 — Decide

Rules-based decision engine with priority weighting and authority checks before any action.

✓ shipped

v0.3 — Act

Approval workflows, action receipts, and supervised execution for real tasks.

✓ shipped

v0.4 — Verify

Outcome tracking that links every action back to the goal it served.

● building

v0.5 — Evolve

Policy evolution from real outcomes, pattern detection, adaptive behavior — agent → partner.

~/orlix/access.sh

$ ./access.sh --request

06Get early access

Orlix is in active development. We're building infrastructure for a true human-AI partnership — and looking for early operators to help shape it.