Walleye

Walleye is a LanceDB server with a write-ahead log on S3 and a thin RAM/NVMe cache, reached through the stock LanceDB remote protocol. We run it for you as a managed instance: you create one, get a URL and a token, and point the LanceDB Python or JavaScript SDK at it.

There is nothing else to operate: no Workers, no schedulers, no shared cluster. Every instance is its own Fly app on a private network with its own S3 bucket and its own bearer token.

• Quickstart — Create an organization and an instance, then connect with the LanceDB SDKs.
• What an instance is — Durability, cache, URL, token and states.
• Shapes and cache tiers — Ramp on shared CPUs or Launch on performance CPUs; small, medium, large or xlarge.
• Stop, start and resize — What a drained stop does and when a resize is accepted.
• Billing — Instance rates, storage and egress, and the free trial.

flowchart LR
    S["LanceDB SDK\nx-api-key: token"] --> I["Instance\nhttps://app.fly.dev"]
    I --> C["RAM / NVMe cache"]
    I --> W["Write-ahead log\nS3 bucket"]

Every write is durable in S3 before it is acknowledged. The cache is disposable: stopping an instance checkpoints every table to S3, and starting it again rebuilds the cache from there.