Vector search, under the glass.
Semantic search in one screen: documents become vectors, your query becomes a vector, and cosine similarity ranks what matches — no keyword matching anywhere.
Your query
Ranked results — 13 documents scored
Email/password, OAuth, and session tokens baked in — no auth server to run, no JWT library to wire up.
Short-lived access tokens, silent refresh, and instant revocation when a user logs out everywhere.
Subscribe to any record and get pushed updates the moment it changes. No polling, no refresh button.
Queue writes on flaky connections and merge them back when you are online — your data stays portable and yours.
Upload files, get URLs, serve them fast worldwide — one bucket, zero config.
Resize on the fly and serve responsive images from the same bucket your app already uploads to.
Store embeddings next to your records and run semantic search — no Pinecone, no second bill.
Turn docs into vectors and ask questions in plain English instead of matching keywords.
$6.99 a month per seat. No meters, no bandwidth math, no surprise bill after launch week.
Try the whole backend for a dollar. If it is not worth it, leave — your data files come with you.
Every seat ships as SQLite files you can download, inspect, and move anywhere. No lock-in, ever.
The same calls from Dart, JavaScript, or plain REST — if you can fetch, you can ship.
Nightly snapshots and encryption at rest, automatic on every seat — nothing to configure.
score = (A·B) / (|A|·|B|) — cosine similarity, 0.9988 for the top hit
Show the raw toy vectors (full transparency)
query “How do I log users in?” = [1.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00]
top doc “Auth without the headache” = [1.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.00, 0.05]
[0] auth[1] realtime[2] storage[3] vector[4] pricing[5] sdk[6] ownership[7] securityWhat you are actually seeing: each document's vector was typed by hand so related topics sit near each other. Real embeddings come from a model trained on language — the ranking math is identical, the vectors are just smarter. InvisibleDB ships this capability built in: no Pinecone, no second bill.