Privacy-First Local Ad Matching (Brave's Catalog Model)
Match ads to users without ever letting their data leave the device.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 80%
Brave built an alternative to conventional ad tech that never lets user data leave the browser. Instead of tracking individuals across the web, Brave distributes an anonymous catalog of ads with keywords to every user in a region; the same catalog updates several times a day for everyone. On-device machine learning—support vector machines and Bayesian classifiers—analyzes a user's browsing and search behavior locally to infer coarse interest segments, then matches catalog ads to those segments entirely within the browser. No behavioral data is ever transmitted to an ad server, eliminating the third-party tracking, fingerprinting, and fraud that plague programmatic advertising. Users who opt in keep 70% of the resulting ad revenue as Basic Attention Token, which they can pass along to creators. The model reconciles advertiser demand for audience targeting with genuine user privacy by moving the decisioning, not the data.
Origin
Developed as Brave's alternative to conventional ad tech, first prototyped on Bitcoin, then built into the Basic Attention Token system, now evolving into a zero-knowledge proof system called Themis on Solana.
Core principles
- 01Decisioning happens on-device, not on a server
- 02The same catalog is broadcast to everyone — no individualized tracking vector
- 03Coarse segmentation prevents re-identification
- 04Users capture the majority of the ad revenue value
How to run it
- 1
Build a region-wide ad catalog
Compile ad creatives (images, pages, videos) plus descriptive keywords into a catalog stored in an edge cache; the same catalog is pushed to a large population in each region.
- 2
Distribute identically to avoid tracking
Update and broadcast the catalog to everyone the same way several times a day so downloading it doesn't identify any individual user.
- 3
Run inference locally
Use lightweight, on-device machine learning (SVMs, Bayesian models) on search queries and browsing behavior to infer coarse interest segments without draining battery or needing cloud compute.
Pro tip Simple statistical models are enough — you don't need TensorFlow or a cloud supercomputer to do useful on-device matching.
- 4
Match ads on-device
Perform the actual ad-to-user matching inside the browser itself so no behavioral data ever leaves the device to an ad server.
Watch out Avoid over-segmenting — individually identifying attributes like zip code, gender, and birthdate can be enough to re-identify a single user.
- 5
Share revenue directly with users
Pay roughly 70% of ad revenue to the user in Basic Attention Token, letting them keep it or forward it to the creators they support.
In the wild
A brand that has lost the ability to retarget ad-blocking users on Chrome runs a campaign on Brave's private ad network instead, targeting broad interest segments rather than individuals.
→ The brand reaches a previously unreachable, ad-blocking-savvy audience while no personal data is exposed, and users earn BAT for their attention.
Common mistakes
Treating 'network' like conventional ad networks
Conventional ad networks track users before deciding which ad to show; replicating that model even with good intentions reintroduces the privacy failure this system is meant to solve.
Ignoring re-identification risk in fine segmentation
Precise, individually addressable targeting provides enough bits of information to fingerprint a single user, similar to how zip code, gender, and birthdate can re-identify someone.
Is it for you?
Best for
Product teams building privacy-preserving monetization
Not ideal for
Advertisers requiring individual-level retargeting
From the transcript
“we put all the ads into a catalog through a link to the creative for the ad the image or web page or video that's…”
“we do the decisioning in the browser based on this catalog so all the machine learning on the mother of all data feeds i mentioned…”
From the episode
The Browser (with Brendan Eich, Chief Architect of Netscape + Mozilla and CEO of Brave)
Brendan Eich