How gorse-io/gorse works
Gorse imports user, item, and feedback data, trains recommendation models on the master, computes offline recommendations on workers, and serves REST requests through server nodes. The README documents recommendation sources including latest, collaborative filtering, user-to-user, item-to-item, and LLM-based approaches. The graph combines those documented capabilities with source-index references and excerpts; several subsystem links are documented at a high level because their implementation files were not sampled.
Main entry points: recommend.go Recommendation logic · tasks.go Training tasks · pipeline.go Offline pipeline · master.go Master service
Components 20
Cluster services
- Master service
master/master.go - Training tasks
master/tasks.go - Worker service
worker/worker.go - Offline pipeline
worker/pipeline.go - REST server
server/server.go - REST endpoints
server/rest.go
Recommendation
- Recommendation logic
logics/recommend.go - Item-to-item
logics/item_to_item.go - Collaborative filtering
model/cf/model.go - CTR model
model/ctr/fm_xla.go - Tensor operations
common/nn/tensor.go - Vector index
common/ann/hnsw.go
Data and storage
- Data store
storage/data/database.go - Recommendation cache
storage/cache/database.go - Vector store
storage/vectors/database.go - Blob storage
storage/blob/blob.go - Model metadata
storage/meta/database.go
Admin and interfaces
- Admin CLI
cmd/gorse-cli/main.go - Admin client
client/admin.go
Other components
- API consumer
Connections 23
- API consumer → REST endpoints: requests recommendations
README.md - REST endpoints → REST server: served by
server/server.go - REST server → Master service: syncs metadata
server/server.go - Master service → Training tasks: runs training
README.md - Worker service → Master service: fetches metadata
worker/worker.go - Worker service → Offline pipeline: runs recommendations
worker/worker.go - Offline pipeline → Recommendation logic: uses recommenders
worker/pipeline.go - Training tasks → Recommendation logic: generates recommendations
master/tasks.go - Training tasks → Collaborative filtering: trains model
master/tasks.go - Training tasks → CTR model: trains model
master/tasks.go - Training tasks → Data store: reads training data
master/tasks.go - Training tasks → Model metadata: stores model metadata
master/tasks.go - Offline pipeline → Recommendation cache: updates recommendations
worker/pipeline.go - Offline pipeline → Data store: loads items
worker/pipeline.go - Recommendation logic → Data store: reads recommendation data
logics/recommend.go - Recommendation logic → Recommendation cache: reads cached results
logics/recommend.go - Recommendation logic → Vector store: uses vector storage
logics/recommend.go - Recommendation logic → Item-to-item: supports item similarity
- Recommendation logic → Vector index: supports nearest neighbors
- CTR model → Tensor operations: uses tensor operations
- Worker service → Blob storage: loads model artifacts
worker/worker.go - Admin CLI → Admin client: calls admin API
cmd/gorse-cli/main.go - Master service → Data store: connects configured store