>
PLAY00:00:00
SLAVA SOLODKIY CUSTOMER FIELD NOTE → OLARES What if? --:--
FIELD NOTES FROM AN OLARES ONE USER · LONDON, SHOREDITCH · 2026

I EXPECTED
A DEVICE.
I GOT AN
EXPERIENCE.

I thought I was buying a powerful local-AI box. After living with it, I realised I had bought something much more valuable to me: a way to discover what I actually wanted personal AI computing to become.

RE_OLARES METASTATE
01 / THE THING I DIDN'T EXPECTHead of Curiosity

THE DEVICE WAS THE PURCHASE. THE EXPERIENCE BECAME THE ASSET.

I have other AI machines in my little garage. Some are faster at particular jobs. Some are cheaper. Some have better hardware. But they are still devices. Olares became a place where I learned by using, breaking, connecting, comparing and trying again. That difference is why this page exists.

Other machines: RAM · VRAM · TOPS · ports · price · benchmarks.
Olares: discover → install → break → learn → connect → automate → imagine.

02 / THINGS I STARTED IMAGININGWhat if?

This is the part of the page that matters most to me. Not a strategy deck. Not a business plan. Just the wishlist that formed in my head as a user — mapped to the Olares products that made each thought appear.

UPDATE · 25 SEP 2026: after I shared this page, Peng Peng, Olares' founder — who, wonderfully, actually reads customer notes himself :) — told me that several of these directions are already close to Olares' roadmap. I am keeping the ideas in their original customer-wish form and only adding status markers — so I do not accidentally present Olares' roadmap back to the team as my idea.
[IDENTITY]

LarePass → digital identity

PARTLY IN MOTION
LarePass Digital Identity

I don't see LarePass as just another sign-in companion. As a customer, I keep wishing it would grow into a fuller identity layer for people, organisations, devices and eventually agents: access, delegation, recovery, trust and proofs — not only login.

Peng / 25 Sep: Olares plans to further define Olares ID in 1.12.8, with DID, payments and reputation based on Olares ID + Olares Domain intended as reusable developer components. My broader person / organisation / device / agent layer is still my own extrapolation.

[EXPERIENCE]

OlaresOS + Market → on other devices

ALREADY IN MOTION
OlaresOS Olares.Market Olares everywhere

If the magic is the experience, why should it live on one box only? I want an official, opinionated way to put OlaresOS + Market on at least some compatible third-party hardware and turn another device into more of the same Olares experience.

Peng / 25 Sep: Olares OS already runs on NVIDIA DGX Spark; DGX Spark can cluster with Olares One; AMD Ryzen AI Max+ 395 has basic support; Mac support is planned via virtual-node technology.

[COMPUTE]

Olares Compute Pod

MY WISH TIINY ENDPOINT: ALREADY POSSIBLE
Olares hardware peripherals

Not just an eGPU (I’m basically imagining an Olares-native version of something like the GIGABYTE AORUS AI Box). A first-class chunk of extra compute that Olares understands, schedules and explains without making me think in PCIe lanes before breakfast.

Partnership and bundle with tiiny?

Peng / 25 Sep: Tiiny should already be able to connect to an Olares One and be configured as an endpoint in Router; deploying Market applications directly onto Tiiny is not there yet.

[LATER]

Spare compute becomes useful

RABBIT HOLE
Olares hardware OlaresOS Olares DePIN

Personal cluster → team compute → company fleet → trusted external capacity. I like the DePIN destination much more when it emerges from something useful first, before it becomes a token story.

[I'D BUY]

Olares × Nothing ?

MY WISH DEVICE SDK: IN MOTION
Olares hardware Co-brand(s)

This is not my most technical idea:) It is my most visceral customer idea. I would buy a transparent, design-forward Olares × Nothing edition because the object itself would finally look as unusual as the experience feels.

Nothing Phone as an Olares console? I can imagine a tiny Linux sandbox, instant secure access to my Olares, and voice-first control built into the phone. Not “AI phone” as another chatbot button — the phone becomes the natural remote control for my personal AI computer.

Olares × Revolut \ SumUp \ World ID \ Octopus \ etc?

Peng / 25 Sep: Olares is already looking for IoT partners and wants a simple SDK for manufacturers of phones, cameras, speakers and headphones. Nothing specifically is still just my customer fantasy.

[Geopolitical Question]

Sovereign AI Gateway

MY WISH
OlaresOS Local AI

Communities powered by Olares: One AI computer, many people. A small company, studio or lab should be able to share serious local AI without sharing one messy login. Roles, quotas, shared models and private workspaces feel native to the idea.

The main question isn't about 'cloud' vs 'local' AI: keep connecting AI's cloud-UX with local-powerhouse - but feel independent. A local Olares box between sensitive company data and outside AI: decide what stays local, what gets redacted, which model runs, and whether an action is allowed.

Think about business client with huge customer base: selling or leasing end-users your co-branded AI mini-station powered by Olares you could create your own corporate 'decentralized AWS' (and better to pay for your customers, rather than to Amazon, no?).

[QUESTION]

An agent can buy compute

RABBIT HOLE RELATED ID / PAYMENTS: IN MOTION
LarePass Digital Identity x Payments Olares DePIN

Eventually an agent will not only use compute; it will buy it. Then my favourite question appears: who authorised the spend? I am deliberately leaving the rabbit hole there for now.

Peng / 25 Sep: Olares ID / DID / payments / reputation are on the product path. Agent spending authority is not something I am claiming Olares has announced — that remains my rabbit hole.

03 / THINGS THIS BOX MADE ME TRYChief Believer

Not achievements. Not a portfolio. Just a few rabbit holes this box nudged me into. I am not trying to impress anyone here as an engineer. I am trying to show what kind of curiosity the experience unlocked in me. One of them predates this whole conversation: I started digging into LarePass back in April 2026.

[LOCAL INFERENCE]

Can I run something that feels too big for one box?

I wondered…

how far local inference could go before the box simply said “no”.

So I tried…

playing with FreeToken and seeing how far a curious non-engineer could push local model workflows.

What stuck with me…

the interesting part was not raw performance. It was the feeling that the machine was becoming a place to explore possibility, not just a spec sheet.

open rabbit hole →
[MODEL ≠ AUTHORITY]

Can a model suggest — while something deterministic stays in charge?

I wondered…

whether a local model could help, without being the thing that is finally allowed to decide.

So I tried…

Nemotron / Qwen + FOL: the model proposes, deterministic checks and formal logic verify, and the model explains afterwards.

What stuck with me…

maybe the real boundary is not local vs cloud. Maybe it is model vs authority.

open rabbit hole →
[BOUNDARIES]

What happens when an agent gets boundaries, not just more intelligence?

I wondered…

what governed local AI feels like in practice: files, network, permissions, approval, audit.

So I tried…

NemoClaw and other agent-style workflows inside Olares, watching what the system allowed, denied or made awkward.

What stuck with me…

the product opportunity may be less “smarter agent” and more “better boundary”.

open rabbit hole →
[TINY FLEET]

What happens when one personal AI computer becomes a tiny network?

I wondered…

whether one box stays one box forever — or quietly wants to become part of a small fleet.

So I tried…

linking Olares with other devices in my garage: Minisforum, Morefine, Tiiny, eGPU paths, separate roles and experiments.

What stuck with me…

this is where thoughts about placement, roles, heterogeneous compute and shared capacity started appearing by themselves.

open rabbit hole →
[LOCAL DECISION / CLOUD RECEIPT]

Can the decision stay local while the cloud only gets a receipt?

I wondered…

whether sovereign AI might mean keeping authority local, while only a lightweight proof leaves the box.

So I tried…

a small trail of experiments around local judgment, external evidence and governed execution.

What stuck with me…

the most interesting future may not be offline-only or cloud-only, but local authority with selective outward proof.

open rabbit hole →
[APR 2026 · IDENTITY]

What if identity is the layer before payment?

I wondered…

what LarePass could become beyond login — long before this Olares job conversation existed.

So I tried…

building a public LarePass-inspired research/reference system around DID / verifiable credentials, onboarding, delegated identity and financial-service extensions — while explicitly separating what I observed from what I invented.

Why it feels relevant again…

ByteTrade now has Otomic for agentic payments and Gaia for decentralized inference. That brings me back to the piece before payment or compute: who authorised this agent to spend, for what, up to what budget, until when, and what evidence survives afterwards?

What stuck with me…

compute can be available and payment can be possible — but neither answers whether the action was authorised. This rabbit hole later evolved into my current work on delegated agent authority.

April 2026 LarePass rabbit hole →
04 / WHAT I WISH OLARES DID FOR MELearning-by-Trying
W7Install an outcomeLet me ask for a capability, not assemble apps + models + endpoints + permissions by hand.
W6See my whole labOlares One · Minisforum · Tiiny · eGPU · NAS — one view of what can run where and why.
W5Make trust visibleWho owns this device? Which agent can touch what? What did I allow once, always, or never?
W4Make Space my control roomHealth · backup · models · users · permissions · recovery · policy across more than one machine.
W3Treat knowledge as infrastructureFirst-class read-only mirrors and connectors for the personal data my agents should understand.
W2Explain failure like a productIf an app cannot install, tell me which layer disagrees and what Olares can safely fix for me.
W1Let me grow without becoming DevOpsThe more ambitious my setup becomes, the less I want the UX to punish me for being curious.
05 / MY ONE TEST(Stay hungry) Stay foolish
Q: IF I REPLACE OLARES WITH AN ORDINARY WINDOWS MINI-PC, DO I STILL WANT THE IDEA?
YES → maybe I just want a computer.
NO → something uniquely Olares-shaped is hiding there.
06 / WHAT I'M TRYING IN THE LABEntrepreneur-in-Residence
  1. Agent permissions

    Give an agent public/private files, local/external models and network rules. Watch what it attempts, what gets denied and where human approval actually belongs.

  2. Heterogeneous placement

    Run the same useful workload across different machines and accelerators. I want Olares to tell me where it belongs rather than force me to become the scheduler.

  3. Mini fleet

    A few devices, a few users, shared models, backup, policy and remote recovery. At what point does a “personal cloud” quietly become a control plane?

  4. Market coherence

    Keep collecting the awkward cases where one-click stops being one-click. Those failures are probably the best product specification Olares could ask for.

07 / CLOSINGOlares.Metastate

I DON'T NEED OLARES TO BECOME ALL OF THESE THINGS. I LIKE IT BECAUSE USING IT KEEPS MAKING ME ASK QUESTIONS. BETTER QUESTIONS.

Most devices in my garage remain devices. Olares became an experience: a place to discover what local AI is useful for, what breaks, what should become one-click, what should stay private, and what new products suddenly become imaginable. For me, that experience is worth more than the box.

Thank you for inspiring me, Slava CUSTOMER · CURIOUS USER · STILL LEARNING · KEEP ASKING Solodkiy.cv