Supabase + pgvector for AI Agents: Memory, RAG, and Auth in One Backend (2026)

TL;DR — Supabase = managed Postgres + pgvector + RLS + edge functions. For AI agent memory and RAG it's the lowest-friction stack in 2026 — five tables, one RPC, and your LangChain agent has long-term memory.

What you'll build

A Supabase project with a documents table for RAG, an agent_memories table for per-user long-term memory, an RLS policy that scopes both to auth.uid(), and a match_documents RPC that LangChain.js calls directly.

Schema

CREATE EXTENSION IF NOT EXISTS vector;

CREATE TABLE documents (
  id BIGSERIAL PRIMARY KEY,
  user_id UUID NOT NULL REFERENCES auth.users (id) ON DELETE CASCADE,
  content TEXT NOT NULL,
  metadata JSONB DEFAULT '{}',
  embedding vector(1536) NOT NULL,
  created_at TIMESTAMPTZ DEFAULT now()
);

CREATE INDEX documents_hnsw_idx ON documents
  USING hnsw (embedding vector_cosine_ops) WITH (m = 16, ef_construction = 64);

ALTER TABLE documents ENABLE ROW LEVEL SECURITY;

CREATE POLICY user_owns_documents ON documents
  USING (user_id = auth.uid())
  WITH CHECK (user_id = auth.uid());

Architecture

flowchart LR
  USER[User in app] --> SB[Supabase Auth JWT]
  SB --> PG[(Postgres + pgvector)]
  PG --> RLS[RLS scopes to auth.uid]
  AGENT[LangChain agent] --> RPC[match_documents RPC]
  RPC --> PG
  PG --> RESULTS[Top-k user docs]
  RESULTS --> AGENT

Step 1 — Create the match RPC

CREATE OR REPLACE FUNCTION match_documents(
  query_embedding vector(1536),
  match_count int DEFAULT 5,
  filter jsonb DEFAULT '{}'
)
RETURNS TABLE (id bigint, content text, metadata jsonb, similarity float)
LANGUAGE plpgsql STABLE
AS $$
BEGIN
  RETURN QUERY
  SELECT d.id, d.content, d.metadata,
         1 - (d.embedding <=> query_embedding) AS similarity
  FROM documents d
  WHERE d.user_id = auth.uid()
    AND d.metadata @> filter
  ORDER BY d.embedding <=> query_embedding
  LIMIT match_count;
END;
$$;

Step 2 — Wire into LangChain.js

import { SupabaseVectorStore } from "@langchain/community/vectorstores/supabase";
import { OpenAIEmbeddings } from "@langchain/openai";
import { createClient } from "@supabase/supabase-js";

const supabase = createClient(
  process.env.SUPABASE_URL!,
  process.env.SUPABASE_ANON_KEY!,
  { global: { headers: { Authorization: `Bearer ${userJwt}` } } },
);

const store = new SupabaseVectorStore(new OpenAIEmbeddings(), {
  client: supabase,
  tableName: "documents",
  queryName: "match_documents",
});

const docs = await store.similaritySearch("appointment policy", 5);

The user's JWT carries auth.uid() — RLS plus the RPC keeps results scoped automatically.

Step 3 — Edge function for ingestion

// supabase/functions/ingest/index.ts
import { OpenAI } from "https://esm.sh/openai@4";
import { createClient } from "https://esm.sh/@supabase/supabase-js@2";

Deno.serve(async (req) => {
  const { content } = await req.json();
  const oai = new OpenAI({ apiKey: Deno.env.get("OPENAI_API_KEY") });
  const v = (await oai.embeddings.create({
    model: "text-embedding-3-small", input: content,
  })).data[0].embedding;

  const sb = createClient(
    Deno.env.get("SUPABASE_URL")!,
    Deno.env.get("SUPABASE_SERVICE_ROLE_KEY")!,
  );
  const userId = (await sb.auth.getUser(req.headers.get("Authorization"))).data.user!.id;
  await sb.from("documents").insert({ user_id: userId, content, embedding: v });
  return new Response("ok");
});

Step 4 — Add agent memory

CREATE TABLE agent_memories (
  id BIGSERIAL PRIMARY KEY,
  user_id UUID NOT NULL,
  agent TEXT NOT NULL,
  kind TEXT NOT NULL,  -- 'episodic' | 'semantic'
  content TEXT NOT NULL,
  embedding vector(1536) NOT NULL,
  created_at TIMESTAMPTZ DEFAULT now()
);
ALTER TABLE agent_memories ENABLE ROW LEVEL SECURITY;
CREATE POLICY uo ON agent_memories USING (user_id = auth.uid());

Step 5 — Realtime updates

Supabase realtime auto-streams changes. Enable on agent_memories and clients see new memories live.

Step 6 — Cost guard

Embedding API costs add up. Hash content before embedding to avoid duplicates:

import crypto from "node:crypto";
const hash = crypto.createHash("sha1").update(content).digest("hex");
// query documents WHERE metadata->>'hash' = hash before embedding

Pitfalls

• Service-role key in browser — never. Use anon key + RLS.
• auth.uid() returns null — RPC was called with anon JWT. Pass the user's JWT in headers.
• Index built on empty table — fine, but rebuild after first 100k inserts for better recall.
• Realtime row size — Supabase realtime caps at 1MB per row.

CallSphere production note

Explore a live demo and compare current plans to find the right fit for your business.

FAQ

Q: Free tier limits? 500 MB DB + 2 GB egress + 50k auth users. Fine for prototypes.

Q: How big can pgvector grow on Supabase? Pro plan + dedicated compute scales to 100M+ rows.

Q: Can I bring my own LLM? Yes — edge functions can call any provider; no Supabase lock-in.

Q: What about data residency? Supabase supports US, EU, AP regions; pick at project create time.

Q: Migrate off Supabase later? pg_dump + restore to any Postgres. No proprietary surfaces beyond auth + storage.

Sources

• Supabase docs — pgvector
• Supabase docs — AI & Vectors
• DEV — Persistent memory with pgvector + Supabase
• DEV — LangChain + Supabase Vector Store