Meter what the agent actually does.

Writes are counted per document, not per byte. Each row you ingest is its own write, so an order counts as one write and each of its line items as another. When a source record is later updated, modified, or deleted, that change flows in through CDC as another write. Sending a message to an agent is a write, and generating embeddings counts as writes too.

No. Reads, queries, and the indexes Quill maintains internally are never metered. You are only charged for writes that matter to you, not the infrastructure writes happening under the hood.

No. A WRU is counted per document regardless of size, so a small record and a large one each count as one write. How fast you ingest does not change the count either.

No. We chose writes over tokens because token usage is unpredictable and differs between models. Quill does not meter or charge for LLM tokens. Those you pay your model provider directly, and nothing at all if you run a model on your own hardware.

Yes. You choose exactly which tables, columns, and rows Quill syncs, so only writes to the data you care about are counted. A high-churn table you do not need, like application logs, can be left out entirely and never costs you a WRU.

Yes, but the 90-day free trial covers that first ingestion, so the migration effectively costs you nothing. The initial import writes a document for each record, so it does consume WRUs; you can load your data and see the value before anything is billed.

WRU is charged in 1M-unit increments, and you can watch your usage live in the Quill dashboard. Whether your usage skews toward many agent messages or heavy data churn, WRU tracks the value you are getting from the system.