n8n-nodes-minimemory

An n8n community node for embedded vector database operations - 100% serverless.

Perform vector similarity search directly within n8n without any external server. Perfect for:
- RAG (Retrieval Augmented Generation)
- Semantic search
- Recommendations
- Deduplication

Features

- No server required - Everything runs locally within n8n
- Similarity search - Find the k most similar vectors
- Hybrid search - Combine vector + keyword (BM25) search
- Multiple metrics - Cosine, Euclidean, Dot Product
- Metadata support - Associate additional information with each vector
- Metadata filtering - MongoDB-style query operators
- Persistence - Save/load from JSON files
- Bulk insert - Insert multiple vectors from previous nodes

Installation

$3

1. Go to Settings > Community Nodes
2. Search for n8n-nodes-minimemory
3. Click Install

$3

``bash
cd ~/.n8n/nodes
npm install n8n-nodes-minimemory
`

Then restart n8n.

Operations

$3

Create a new vector database in memory.

| Parameter | Description | Example |
|-----------|-------------|---------|
| Database Name | Unique name for the DB | my_vectors |
| Dimensions | Number of dimensions | 384 (MiniLM), 1536 (OpenAI) |
| Distance Metric | Similarity metric | cosine, euclidean, dot |
| Index Type | Index type | flat (exact), hnsw (fast) |

$3

Insert a single vector with ID and optional metadata.

| Parameter | Description |
|-----------|-------------|
| Vector ID | Unique identifier |
| Vector | Array of numbers [0.1, 0.2, ...] |
| Metadata | JSON object {"title": "Doc 1"} |

$3

Bulk insert vectors from input items.

| Parameter | Description | Default |
|-----------|-------------|---------|
| ID Field | Field containing the ID | id |
| Vector Field | Field containing the vector | embedding |
| Metadata Fields | Fields for metadata (empty = all) | |

Example input:
`json
[
{"id": "doc1", "embedding": [0.1, 0.2, ...], "title": "Document 1"},
{"id": "doc2", "embedding": [0.3, 0.4, ...], "title": "Document 2"}
]
`

$3

Search for the k nearest neighbors using vector, keyword, or hybrid search.

| Parameter | Description | Default |
|-----------|-------------|---------|
| Search Mode | vector, keyword, or hybrid | vector |
| Query Vector | Vector for similarity search | (required for vector/hybrid) |
| Keywords | Text for BM25 keyword search | (required for keyword/hybrid) |
| Text Fields | Metadata fields to search | content,text,title,description |
| Number of Results (K) | Number of results | 10 |
| Include Vectors | Include vectors in result | false |
| Minimum Similarity | Filter by minimum similarity | 0 |
| Use Metadata Filter | Enable metadata filtering | false |
| Metadata Filter | JSON filter (see below) | {} |
| Hybrid Alpha | Vector(1) vs Keyword(0) balance | 0.5 |
| Fusion Method | rrf or weighted | rrf |
| BM25 K1 | Term saturation parameter | 1.2 |
| BM25 B | Length normalization | 0.75 |

Output:
`json
{
"success": true,
"searchMode": "hybrid",
"filterApplied": true,
"results": [
{"id": "doc1", "score": 0.032, "vectorSimilarity": 0.9, "keywordScore": 12.4, "metadata": {...}},
{"id": "doc2", "score": 0.028, "vectorSimilarity": 0.8, "keywordScore": 8.2, "metadata": {...}}
]
}
`

Metadata Filtering (New in v0.2.0)

Filter search results by metadata fields using MongoDB-style operators.

$3

`json
// Exact match
{"category": "tech"}

// Multiple conditions (implicit AND)
{"category": "tech", "author": "John"}
`

$3

| Operator | Description | Example |
|----------|-------------|---------|
| $eq | Equal (implicit) | {"status": "active"} |
| $ne | Not equal | {"status": {"$ne": "deleted"}} |
| $gt | Greater than | {"score": {"$gt": 0.5}} |
| $gte | Greater or equal | {"price": {"$gte": 100}} |
| $lt | Less than | {"age": {"$lt": 30}} |
| $lte | Less or equal | {"count": {"$lte": 10}} |

$3

| Operator | Description | Example |
|----------|-------------|---------|
| $in | In array | {"category": {"$in": ["tech", "science"]}} |
| $nin | Not in array | {"type": {"$nin": ["spam", "ad"]}} |

$3

| Operator | Description | Example |
|----------|-------------|---------|
| $contains | Contains (case-insensitive) | {"title": {"$contains": "AI"}} |
| $startsWith | Starts with | {"name": {"$startsWith": "Dr."}} |
| $endsWith | Ends with | {"file": {"$endsWith": ".pdf"}} |

$3

| Operator | Description | Example |
|----------|-------------|---------|
| $exists | Field exists | {"email": {"$exists": true}} |

$3

`json
// AND - all conditions must match
{
"$and": [
{"category": "tech"},
{"score": {"$gt": 0.8}}
]
}

// OR - any condition matches
{
"$or": [
{"type": "article"},
{"type": "blog"}
]
}
`

$3

Access nested object fields using dot notation:

`json
{"user.profile.country": "US"}
`

$3

Find tech articles with high score:
`json
{
"$and": [
{"category": "tech"},
{"score": {"$gte": 0.8}}
]
}
`

Find documents from specific authors:
`json
{"author": {"$in": ["Alice", "Bob", "Charlie"]}}
`

Find recent documents with keyword:
`json
{
"timestamp": {"$gt": "2024-01-01"},
"title": {"$contains": "machine learning"}
}
`

$3

Get a vector by its ID.

$3

Delete a vector by its ID.

$3

Save the database to a JSON file.

$3

Load a database from a JSON file.

$3

Get database information and statistics.

$3

List all databases in memory.

$3

Remove all vectors from the database.

$3

Remove a database from memory.

Example Workflow

$3

`
[Trigger]
|
[OpenAI Embeddings] -> [Minimemory: Insert Many]
|
[Query Input]
|
[OpenAI Embeddings] -> [Minimemory: Search] -> [OpenAI Chat]
|
[Response]
`

$3

1. Create DB (run once):
- Operation: Create Database
- Database Name: docs
- Dimensions: 1536 (OpenAI)
- Distance: cosine

2. Index documents:
- Connect node that generates embeddings
- Operation: Insert Many
- ID Field: id
- Vector Field: embedding

3. Search:
- Connect query with embedding
- Operation: Search
- Query Vector: {{ $json.embedding }}
- K: 5

Persistence

Databases live in memory while n8n is running. To persist:

`
[Startup Trigger] -> [Minimemory: Load from File]
|
(DB available for use)
|
[Before shutdown] -> [Minimemory: Save to File]
`

Embedding Compatibility

| Model | Dimensions | Notes |
|-------|------------|-------|
| all-MiniLM-L6-v2 | 384 | Fast, good for short text |
| all-mpnet-base-v2 | 768 | Better quality |
| text-embedding-ada-002 | 1536 | OpenAI legacy |
| text-embedding-3-small | 1536 | OpenAI new |
| text-embedding-3-large | 3072 | OpenAI high quality |
| embed-english-v3.0 | 1024 | Cohere |

Performance

- Flat Index: O(n) - Exact, ideal for < 10,000 vectors
- HNSW Index: O(log n) - Approximate, for large datasets

Troubleshooting

$3


Make sure to run Create Database or Load from File first.

$3

The inserted/searched vector has a different number of dimensions than the DB.

$3

1. Verify npm run build completed without errors
2. Check the link with npm ls -g --link`
3. Restart n8n

Support

For issues and feature requests, please visit:
https://github.com/MauricioPerera/n8n-nodes-minimemory/issues

License

MIT