SaaSVerified Nov 5, 2026

Supabase AI app with semantic search

Postgres with pgvector for embeddings, the AI SDK for model calls, and prompt caching to keep costs down.

by Promptu

What you end up with

An app that stores embeddings in Supabase Postgres (pgvector), answers questions with the AI SDK, and reuses cached prompt prefixes. The prompt helps explain slow queries as the data grows.

Build order

1
MCP serverDatabase access for your agent
Supabase MCP server
Connect agents to your Supabase projects: schema, SQL, logs and docs. Scope it to one project and read-only when you can.
2
TechnologyEmbeddings
pgvector
Open-source vector similarity search for Postgres. Keep embeddings next to your data, with no separate vector database.
```
CREATE EXTENSION vector;
```
3
TechnologyModel calls
AI SDK
TypeScript toolkit for building AI features (streaming, tool calling, structured output) across model providers.
```
npm install ai
```
4
TechnologyCost
Prompt caching (Claude API)
Reuse a processed prompt prefix (tools, system, messages) to cut cost and latency on repeated context.
5
PromptPerformance
Why is this Postgres query slow?

Review commands before running them. Promptu never runs code on your machine.