Vector search

Semantic similarity search built into the DB.

Dev-limited

DB.createVectorIndex / findSimilar

🧪

Real calls, placeholder embeddings. createVectorIndex, insertOne and findSimilar all execute against the dev runtime. But there is no embedding model wired here — vectors come from a deterministic hash of the text, so the scores are illustrative, not semantically meaningful. In production you embed text with a real model before insert and query.

Result

Run a query to attempt a real findSimilar call.

Vector calls

No output yet — run the demo above.

Setup flow (real calls)

⚠️
createVectorIndex
createVectorIndex error: Resource limit exceeded: collections limit is 10, current count is 13
✅
insertOne × corpus
5 docs present (0 new)

Corpus · 5 docs · 16-dim

Edge compute run code close to users at the network edge for low latency
Vector search semantic similarity search over embeddings and nearest neighbours
Key-value store fast ephemeral key value cache with time to live expiry
Authentication user login sessions tokens groups and permission policy checks
Realtime channels websocket pub sub presence and live message broadcast