supabaseVectorStore

Supabase-hosted pgvector with direct mutations and a purpose-specific similarity-search RPC.

import { supabaseVectorStore } from '@agentskit/memory'

const store = supabaseVectorStore({
  url: process.env.SUPABASE_URL!,
  serviceRoleKey: process.env.SUPABASE_SERVICE_ROLE_KEY!,
})

@supabase/supabase-js is an optional peer dependency loaded lazily. Keep the service-role key in server-only code.

#Server-side setup

Run this once in the Supabase SQL editor. Change both vector dimensions if your embedding model does not produce 1536-dimensional vectors.

create extension if not exists vector with schema extensions;

create table public.agentskit_vectors (
  id text primary key,
  content text not null,
  embedding extensions.vector(1536) not null,
  metadata jsonb not null default '{}'::jsonb
);

create or replace function public.match_agentskit_vectors(
  query_embedding extensions.vector(1536),
  match_count integer default 10,
  match_threshold double precision default 0,
  filter jsonb default '{}'::jsonb
)
returns table (
  id text,
  content text,
  metadata jsonb,
  similarity double precision
)
language sql
stable
security invoker
set search_path = ''
as $$
  select
    vectors.id,
    vectors.content,
    vectors.metadata,
    1 - (vectors.embedding operator(extensions.<=>) query_embedding) as similarity
  from public.agentskit_vectors as vectors
  where vectors.metadata @> filter
    and 1 - (vectors.embedding operator(extensions.<=>) query_embedding) > match_threshold
  order by vectors.embedding operator(extensions.<=>) query_embedding
  limit least(greatest(match_count, 1), 100);
$$;

revoke all on function public.match_agentskit_vectors(
  extensions.vector,
  integer,
  double precision,
  jsonb
) from public, anon;

grant execute on function public.match_agentskit_vectors(
  extensions.vector,
  integer,
  double precision,
  jsonb
) to service_role;

The adapter writes through PostgREST upsert, deletes through delete().in(...), and calls only match_agentskit_vectors for search. It does not expose an RPC that accepts arbitrary SQL.

#Config

Option	Type	Default
`url`	`string`	required
`serviceRoleKey`	`string`	required (server-side only)
`table`	`string`	`'agentskit_vectors'`
`matchFunction`	`string`	`'match_agentskit_vectors'`
`topK`	`number`	`10`

If you change table, provide a matching purpose-specific function and set matchFunction to its name.

#Surface

Implements VectorMemory: store(docs) / search(embedding, options) / delete(ids). The default SQL function accepts simple metadata equality filters through JSON containment, for example { tenantId: 'acme' }. To support compound or comparison operators, provide a custom bounded RPC with the same parameters and return columns.

#Security

Never expose SUPABASE_SERVICE_ROLE_KEY to a browser or mobile client.
Keep the search function security invoker; it needs no definer privileges.
Do not replace it with a function that accepts SQL text from the caller.
Use a dedicated table and the narrowest database grants required by your server-side role.

#Cleanup

Remove the integration objects if you no longer use them:

drop function if exists public.match_agentskit_vectors(
  extensions.vector,
  integer,
  double precision,
  jsonb
);
drop table if exists public.agentskit_vectors;

pgvector — use this when you already have a safe SQL runner.
Memory overview

Explore nearby

✎ Edit this page on GitHub·Found a problem? Open an issue →·How to contribute →