Walleye docs
Running Walleye yourself
Walleye is one binary. It speaks the LanceDB remote protocol, keeps its write-ahead log in object storage, and caches on local NVMe and in memory. Everything in the examples runs on this binary with nothing managed behind it.
- Configuration, the whole environment surface
- A single node, against any S3-compatible bucket
- A cluster, three nodes with quorum durability
- Budgets, what the node reserves and why
- What you own, and what is different on the managed side
The shortest version
export WALLEYE_BUCKET=my-bucket
export AWS_ACCESS_KEY_ID=...
export AWS_SECRET_ACCESS_KEY=...
export AWS_ENDPOINT=https://t3.storage.dev # any S3-compatible endpoint
export AWS_REGION=auto
walleye-node
It prints a generated token if you did not set one. Point the LanceDB client at it and the rest is ordinary:
import lancedb
# The node's address goes in host_override, never in the URI: the client
# picks its transport from the scheme, and db:// is what means remote.
db = lancedb.connect(
"db://walleye",
api_key="<token>",
host_override="http://127.0.0.1:8080",
region="local",
)
db.create_table("docs", data=[{"id": "a", "vector": [0.1, 0.2]}])