Signal-first
A feed ranked around your real interests
Signal -> Prioritize -> Discover
A feed for your mind, not for the algorithm.
Ocno helps you monitor what matters by ranking sources around your interests instead of generic engagement signals.
A feed ranked by your actual interests
Less noise and more relevant sources
Discover adjacent signals without doomscrolling
0:00 / 0:00
Lower noise
Avoid generic engagement bait and obvious hoaxes
Cross-source
See relevant articles and posts in one stream
Stay current
Monitor topics without endless manual filtering
Join Early Access
Get priority onboarding, founder-circle perks, and direct influence on what we build next.
HOW FEED RANKING WORKS
From browsing signal to a personal feed.
Ocno splits collection from ranking so the system can stay useful without turning the server into a profile of you.
Browse
1
You browse. Ocno listens and reads with privacy in mind.
As you browse, Ocno pays attention to the pages you open and the topics you spend time with. That taste signal stays local-first, so your raw browsing history does not need to become a cloud profile.
Privacy highlight: your personal taste layer stays on your device.
Precompute + Pull
2
Each day OCNO precomputes the full public update set into embeddings.
The server scans the public internet, turns the full day into embeddings, and your device pulls the entire bundle in about 3 seconds. Because you pull the whole day instead of a personalized slice, the server still cannot tell what you like.
Everyone gets the full prepared day. Preference only appears when ranking happens locally.
Rank
3
Your local ranker orders the day according to your taste.
Once the daily bundle is on-device, the local ranker scores it against your reading patterns and interests. The feed you see is shaped by your taste, not by a generic engagement algorithm.
Result: one daily stream, locally ranked for you.
ARCHITECTURE
Everything stays
on your machine.
No data leaves your device. Ever. Here's exactly what talks to what.
Public articles are delivered in bulk
Each day, the OCNO server crawls public news sources and delivers the full article bundle to your device in bulk. Because everything arrives upfront, the server does not know what you actually read or like.
Your local model does the ranking
A local ranker compares that bundle against your own browsing context, memory, and reading profile - entirely on-device.
Personalisation stays local
The server provides the inputs, but it does not see the local signals or ranking logic that shape your feed. The relevance layer lives on your side of the boundary.
FEED
Designed for high-signal discovery.
A good feed should narrow the field, respect your interests, and keep the ranking logic on your side.
01Relevance
Rank what actually matters
- Surface updates aligned with your real interests
- Prioritize signal over generic popularity
- Spend less time filtering and more time learning
- Let curiosity shape ranking instead of engagement loops
02Breadth
Pull from across the web
- Blend multiple public sources into one reading stream
- Track adjacent topics without manual monitoring
- Discover relevant angles you would not have searched for
- Keep research discovery active and wide without losing focus
03Privacy
Keep ranking local-first
- Your interest profile stays on-device
- No dependency on external social ranking systems
- Personalization without exposing your intent map
- A private feed that still feels smart and current
What enabled this?
5-10x Faster
Thanks to its compiler-optimized engine and full GPU utilization, Ocno delivers up to 5–10× higher performance compared to traditional CPU-based in-browser model execution.
Smaller models are catching up 🔥
Smaller models like Qwen3 4B (GPU and 4GB RAM needed) can achieve quality comparable to large earlier models such as GPT-4o in many tasks proven by various benchmarks.
Model Compression & Quantization
Advances in quantization and model compression drastically reduced memory requirements, making powerful models practical on everyday devices. You can now run model comparable to GPT-4o on just 3–4 GB of RAM — enabling capabilities once limited to large cloud systems.
MAKER
Meet the Builder
I replaced my social life with shipping.
