Built-in components - Verglas

Verglas includes prebuilt Durable Objects for common state and data workloads. They use the same placement, isolation, storage, and binding model as your own code.

Component	Use it for	Worker API
Durable Object	Per-key state, coordination, SQL, alarms, WebSockets	`env.NAME.get(id).fetch(...)`
Stream	Durable ordered JSON ingestion	`env.STREAM.send(records)`
Pipeline	Filter, project, fan out, and checkpoint Stream records	Configured SQL
Sink + Catalog	Idempotent Parquet publication to Iceberg	Pipeline destination / Iceberg REST
Query	Durable aggregate read models	`env.QUERY.query(endpoint, params)`
Vectorize	Vector CRUD and similarity search	`insert`, `upsert`, `query`, `queryById`
Graph	Property graph CRUD and bounded traversal	`neighbors`, `shortestPath`

These names describe component roles, not separate infrastructure products. A Stream is a Durable Object specialized for append and read. A Pipeline is a Durable Object specialized for a cursor and retry state. Their capacity follows the same serverless object model.

Pick the smallest component

Start with a Worker for stateless request handling.
Add your own Durable Object when the state model is application-specific.
Add a Stream when events must survive beyond a request.
Add a Pipeline when those events need repeatable transformation and delivery.
Add Vectorize or Graph when the access pattern is similarity or traversal.
Add Query when a low-latency endpoint should read a maintained aggregate.

Bindings use fixed resource names. Code cannot construct an arbitrary Stream, index, graph, or query name at runtime.