A step-by-step guide to using SlimArmor, your personal vector database for semantic search.

SlimArmor lets you store text and search it by meaning, not just keywords.

Example:

• You store: "Dogs are loyal companions"
• You search: "furry pets"
• It finds the dog text, even though the words are different!

This is called semantic search and it's powered by AI embeddings.

Your SlimArmor API lives at:

https://kamenxrider--95fbe492ffe111f0bee942dde27851f2.web.val.run

Open that in a browser to see all available endpoints.

To add text to the database, send a POST request:

curl -X POST https://YOUR_ENDPOINT/upsert \
  -H "Content-Type: application/json" \
  -d '{
    "id": "note-1",
    "text": "The quick brown fox jumps over the lazy dog",
    "meta": {"category": "example"}
  }'

What the fields mean:

• id - Unique identifier (you choose)
• text - The text you want to search later
• meta - Optional extra data (tags, categories, etc.)

To find similar text:

curl -X POST https://YOUR_ENDPOINT/search \
  -H "Content-Type: application/json" \
  -d '{"query": "animals jumping"}'

Response:

{
  "results": [
    {
      "id": "note-1",
      "text": "The quick brown fox jumps over the lazy dog",
      "distance": 0.52
    }
  ]
}

The distance tells you how similar it is:

• 0.0 = Exact match
• 0.5 = Related
• 1.0 = Not related

To only get highly relevant results, add maxDistance:

curl -X POST https://YOUR_ENDPOINT/search \
  -H "Content-Type: application/json" \
  -d '{"query": "animals jumping", "maxDistance": 0.6}'

This filters out anything with distance > 0.6.

Recommended thresholds:

• 0.5 = Very strict (only best matches)
• 0.6 = Balanced (good default)
• 0.7 = Loose (more results)

Not sure what threshold to use? Use the calibrate endpoint:

GET /calibrate?q=your+search+query

It will analyze your data and suggest:

• Tight - Top 3 results only
• Balanced - Top 10 results (use this!)
• Loose - All retrieved results

See how much data you have:

GET /stats

Shows:

• Total records
• Storage used
• Estimated capacity

curl -X POST https://YOUR_ENDPOINT/delete \
  -H "Content-Type: application/json" \
  -d '{"id": "note-1"}'

GET /get?id=note-1

GET /list

curl -X POST "https://YOUR_ENDPOINT/clear?confirm=yes"

import * as db from "https://esm.town/v/kamenxrider/slimarmor/vectordb.ts";

// Setup (run once)
await db.setup();

// Add a record
await db.upsert("my-note", "This is my important note", { 
  tag: "work" 
});

// Search
const results = await db.search("important information", 5, 0.6);
console.log(results);

// Delete
await db.remove("my-note");

Just use the HTTP endpoints with any language that can make HTTP requests.

Use descriptive IDs like blog-post-123 or note-2024-01-15 instead of just numbers.

Shorter, focused text works better than huge documents. If you have long documents, consider splitting them into chunks.

Store extra info in meta so you can filter later:

{
  "id": "doc-1",
  "text": "...",
  "meta": {
    "category": "tech",
    "date": "2024-01-15",
    "author": "john"
  }
}

Use /calibrate?q=... to find the right maxDistance for your use case.

Check /stats periodically. You have ~47,000 records per GB.

Make sure the NEBIUS_API_KEY environment variable is set in Val Town.

Try lowering maxDistance (e.g., 0.5 instead of 0.7).

• Check if you have data: GET /stats
• Try raising maxDistance (e.g., 0.8)
• Try a simpler query

Each insert needs an API call to generate embeddings (~460ms). This is normal.

import * as db from "https://esm.town/v/kamenxrider/slimarmor/vectordb.ts";

// Add some notes
await db.upsert("note-1", "Meeting with Bob about Q4 budget planning");
await db.upsert("note-2", "Research machine learning frameworks for project");
await db.upsert("note-3", "Call mom for her birthday on Saturday");

// Search for work stuff
const workNotes = await db.search("business meetings finance", 5, 0.6);

// Search for personal stuff  
const personalNotes = await db.search("family calls birthdays", 5, 0.6);

1. Check / endpoint for all available endpoints
2. Check /stats to see your data
3. Use /calibrate?q=... to tune your searches

Happy searching! 🔍