Temporal Profiles
Model user interests over time with configurable half-life decay curves and topic affinities.
CactusBrain Personalize / Developer Preview
Personalization without user tracking or tracking databases.
Rank items to one user's behaviour on their own device. No models to train, no central telemetry to collect, and zero user data ever leaving the handset.
LOCAL INTERACTIONSTaps / Views
↓TEMPORAL PROFILEHalf-life decay
↓RANKED CANDIDATESPersonalized list
01 / Implemented SDK behavior
On-Device Ranking
Score hundreds of candidates in sub-millisecond local execution without network calls.
Exploration Modes
Prevent echo chambers with built-in epsilon-greedy and contextual exploration algorithms.
Encrypted Persistence
Profile weights and observation history are stored atomically in encrypted device storage.
02 / Swift API
Immediate feedback loops without backend round-trips.
When a user interacts with content, their feed updates instantly. No batch pipelines, no vector database hosting costs.
import CactusBrainPersonalizeSDK
let personalizer = try await CactusBrainPersonalize.load(
collection: "feed-ranking",
profile: .decaying(halfLife: .hours(48))
)
// Observe local user interactions (opens, clicks, dwell time)
try await personalizer.observe(
itemId: "article-892",
affinity: .positive(weight: 1.0)
)
// Rank candidate items completely on-device
let rankedItems = try await personalizer.rank(
candidates: candidateIds,
mode: .hybrid(exploration: 0.15)
)03 / Privacy by Architecture
GDPR, CCPA, and ATT compliant by default.
Because personalization state stays on the client, you don't need complex consent flows for server-side user tracking. The data never leaves the operating system sandbox.