🔌 Connecting Supabase, Qdrant, and Ollama with Guardian

Now that you’ve got the stack installed and running locally, let’s connect the key services together so Guardian can start making use of them.

This guide assumes:

• Docker containers are running for Supabase and Qdrant
• Ollama is installed locally with llama3 running
• You’ve cloned or initialized a project folder with your Guardian server backend

---

🔹 Step 1: Connect to Supabase

Inside your Guardian server (Hono/Express), you’ll want to connect directly to the Postgres DB exposed by Supabase.

If you’re using Drizzle ORM like I do:

// db.config.ts
import { drizzle } from 'drizzle-orm/node-postgres';
import pg from 'pg';

const pool = new pg.Pool({
  connectionString: 'postgresql://postgres:postgres@localhost:54322/postgres'
});

export const db = drizzle(pool);

📝 Port 54322 is Supabase’s default if you’re using the Docker setup.

---

🔹 Step 2: Connect to Qdrant

You’ll use the Qdrant JS client:

pnpm add @qdrant/js-client-rest

Then initialize:

// qdrant.ts
import { QdrantClient } from '@qdrant/js-client-rest';

export const qdrant = new QdrantClient({
  url: 'http://localhost:6333'
});

You can now create collections, insert embeddings, and run similarity searches.

---

🔹 Step 3: Connect to Ollama

Use the Ollama SDK or just a simple fetch if you’re running locally:

const response = await fetch('http://localhost:11434/api/generate', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({
    model: 'llama3',
    prompt: 'What is Guardian AI?'
  })
});

const result = await response.json();
console.log(result.response);

⚠️ You can wrap this in a utility like runLLM() to abstract model calls.

---

🔹 Step 4: Define Your Agent’s Memory Loop

Here’s the high-level logic:

const result = await runLLM(prompt);

// Embed result
const embedding = await embedText(result.response); // use `@pinecone-database/embedding-openai` or your own

// Store in Qdrant
await qdrant.upsert('guardian-memory', {
  points: [
    {
      id: uuid(),
      vector: embedding,
      payload: {
        type: 'agent_thought',
        source: 'blog-reader',
        content: result.response
      }
    }
  ]
});

---

✅ Final Checks

Make sure:

• Supabase has a working schema (start with tasks, memories, or agents)
• Qdrant is seeded with at least one collection (e.g. guardian-memory)
• Ollama’s model is loaded and not timing out

---

🚀 Next Up

Now that Guardian can:

• Talk to Supabase
• Store/retrieve semantic memory via Qdrant
• Use a local LLM to reason and respond