Last verified against Looki’s public website, Terms of Service, and Privacy Policy on 26 April 2026. Pricing, features, and policies change – check the source if anything looks off.
Ostler: $49.99 + $24.99/mo – runs on hardware you already own; the assistant runs locally.
Looki: $199 hardware + Looki+ Free / $5.99 / $9.99 per month – the assistant runs on AWS, with OpenAI as a documented sub-processor.
Looki is the most interesting wearable AI camera shipping in 2026. A 32-gram pendant with a 12MP sensor and three microphones, captured in short clips, paired with an AI assistant. The architecture forces a trade-off: a pendant that small cannot run a multimodal reasoner, so the reasoning lives in the cloud – specifically, on Looki’s AWS, with OpenAI as a third-party AI provider named in their privacy policy. Their founder, Yang Sun (ex-Google Assistant), is candid about this in a recent founder interview with Robert Scoble: the principle, in his words, is “we give the control to the user,” meaning the user decides what to upload. Once uploaded, the data lives in Looki’s cloud and is processed by Looki’s AI sub-processors. Ostler is the opposite shape – a memory layer that runs on Apple Silicon you already own, with no cloud step at all.
| Ostler | Looki L1 | |
|---|---|---|
| Where the AI runs | On your Mac (local Ollama) | AWS in your region, with OpenAI as sub-processor |
| Third-party AI providers | None | OpenAI (named in Looki’s privacy policy) |
| Where your data lives | Encrypted on your Mac | Buffered on-device, uploaded to AWS for AI features |
| Content licence in their Terms | None – you keep all rights | Perpetual, irrevocable, sublicensable, transferable (Section 6.1) |
| Capture device | macOS app + iOS companion | 32g pendant, always-wearable |
| Capture mode | Explicit, per-conversation | Story Mode (5–11s clips at 30s/1min/2min/3min intervals) |
| Bystander signalling | Phone or laptop in hand – visibly capturing | Trust Light LED, cannot be disabled |
| Hardware cost | Reuses your existing Mac | $200 early bird, $250 regular |
| Battery life | N/A – runs on mains-powered Mac | ~12 hours per charge (founder: "won’t get me through the middle of the night") |
| Public API for third-party agents | Local hub API, Mac-side | Yes – users pipe captured data into OpenClaw and other cloud agents |
| Auto meeting summaries | Yes – via local Whisper + the conversation pipeline | Yes – cloud-routed; AI mode auto-switches to audio capture in meetings |
| Compliance certifications | Independent security audit commissioned (Lester Lim / S-RM) | SOC 2 + GDPR certifications stated as in-progress, not yet attained |
| Subscription model | $24.99/mo (replaces other AI subs) | Free / $5.99 / $9.99 per month tiers |
| Storage cap | Your local disk | 30GB / 500GB / 1TB depending on tier |
| Multi-platform import | 20 GDPR exports (LinkedIn, WhatsApp, etc.) | "Coming soon" on Prime tier only |
| Relationship intelligence | Local – warmth, reciprocity, drift detection | Cloud-routed via OpenAI |
| Personal wiki | Auto-generated, 21 page types | No |
| Auto-generated daily vlog | No | "1 Day in 60 Seconds" recap |
| Works offline | Yes | AI features require cloud |
| Funding stage | Bootstrapped – one founder, HK | $20M+ Series A (Jan 2026, Ant Group / Meituan Dragon Pearl); Looki Tech Limited, Toronto |
Looki is hardware-first and well-funded. The team is serious – CEO Yang Sun was a founding member of Google Assistant; CTO Liu Bocong led autonomous-driving algorithms at Meituan-W. The form factor is genuinely beautiful and the Trust Light is a thoughtful answer to the bystander problem. None of that is in dispute.
What is in dispute is where the assistant runs. A 32-gram pendant on a Snapdragon W5 Gen 2 cannot host a multimodal reasoner – the silicon, the battery, and the thermal envelope all rule it out. So the wearable captures, optionally filters on-device, then the heavy lifting happens on Looki’s AWS. That is the only architecture that lets a pendant exist in 2026. It is a real engineering achievement. It is also a structural privacy ceiling.
Looki publishes reasonable on-device protections: footage stored locally first, manual upload approval, encryption at rest, regional AWS storage, a hardware Trust Light that cannot be turned off. Those are better than the wearable-AI baseline. The catch is in the part of the system that does the actual reasoning: the moment you ask the AI assistant a question about your day, your content has to leave the device, because the assistant is on the cloud. Looki’s privacy policy names OpenAI as a third-party AI provider for that step. So the data path for any AI-powered feature is L1 → Looki app → AWS → OpenAI.
Looki commits to not training their own algorithms on your content and requires their AI sub-processors to abide by the same. We have no reason to doubt that. The point is that the trust model shifts from architectural to contractual: we promise not to misuse what we hold rather than we cannot misuse what we never receive. Section 6.1 of their Terms then takes a "worldwide, non-exclusive, perpetual, irrevocable, royalty-free, fully paid, sublicensable and transferable license" over user content. That language is standard for cloud platforms; it is the inverse of where Ostler sits, where the user keeps everything because nothing is uploaded.
Ostler holds a different position. The reasoning model runs on your Mac. The encrypted databases sit on your Mac. There is no AWS region, no OpenAI hop, no perpetual content licence in the contract. The privacy story is architectural, not contractual.
Looki is fundamentally a camera that adds an assistant. It generates new content – clips of your day – and reasons about that. Ostler is a memory layer that ingests content you already own: twenty years of messages, emails, photos, and contacts you have already exported from existing platforms. Different surface, different threat model. Looki creates novel recordings of bystanders; Ostler reasons over data you already have legitimate possession of.
Both shapes are valid. They are answers to different questions. Looki asks “what if I never had to remember anything ever again?”. Ostler asks “what if everything I have already remembered were searchable, connected, and mine?”.
In Yang Sun’s interview with Scoble he stated their privacy posture clearly: “everything it captures will be stored on device by default... only getting the user consent and the user might explicitly click on this button upload and analyze, all the files will leave the device and then transfer to the cloud and the AI will start to analyze.” That is an honest statement of the architecture: local until you upload, cloud once you do, and the AI features require the upload. He also confirmed Looki is pursuing SOC 2 and GDPR certifications – which, in our reading, is the right next step for a vendor whose trust model is contractual rather than architectural. Two products, two coherent positions.
Two specifics from the same interview that are worth knowing if you are choosing between them. First, Looki ships a public API and explicitly supports piping captured content into OpenClaw and other agent platforms – meaning a Looki user’s data path in the agent-integrated case is L1 → Looki cloud → OpenAI → OpenClaw cloud → back. The cloud hops compound. Second, Looki’s “AI mode” auto-detects context and switches the camera off in favour of audio recording during meetings, then auto-generates a meeting summary on the cloud – useful, novel, and a direct overlap with Ostler’s remote-conversation pipeline, except that ours runs the audio through local Whisper on your Mac, not anyone’s cloud.
Looki has hardware and outputs Ostler does not. The 32g pendant is a real engineering achievement; the auto-generated “1 Day in 60 Seconds” vlog and the comic-page output are creative features Ostler has not built. Their AI assistant attempts behavioural and lifestyle coaching – calorie tracking, sleep nudges, patience feedback – based on what the camera captures. Those overlap with the territory Ostler claims to serve, and we should not pretend otherwise.
The difference is the trust posture, not the feature set. Looki’s coaching reads your content from AWS, via OpenAI, under a Terms section that grants them broad use rights. Ostler’s coaching reads your content from a SQLite file on your own SSD that nothing else can see. Both can answer “am I over my calorie limit today?”. Only one of them does it without your day’s footage passing through a third-party AI vendor.
What Looki cannot match without rebuilding the company is the architecture of trust: a system that holds nothing on a vendor server because there is no vendor server. Their unit economics depend on cloud subscriptions and cloud inference, the way every other wearable does. Ours do not. That is a structural difference that survives whatever pricing or feature changes either side ships. Looki’s “3rd-Party Media Import”, listed as “Coming soon” on the Prime tier, will eventually move them toward Ostler’s multi-platform-import territory – but the data flow into AWS and OpenAI stays the same.
Do you want a beautiful pendant that captures your day and reasons about it on AWS, with OpenAI as a documented sub-processor? Or do you want a memory system that reasons about your existing life, on hardware you already own, with no off-device step? Looki is the better wearable. Ostler is the better memory layer. If you already have a Looki, Ostler is what would make its transcripts – and your other twenty years of data – cohere without the cloud round-trip.