chromadb

name	chromadb
description	Semantic search for model descriptions and content using ChromaDB vector database. Use when searching for similar models, finding related content, querying vector collections, or looking up models by description or semantic similarity.

ChromaDB Skill

Semantic search and vector storage using ChromaDB for finding similar content across collections.

Requirements

pip install chromadb

Usage

Query for similar items

python3 {baseDir}/query.py query "gothic fantasy creature" --collection models [--limit 5]

Find similar to existing item

python3 {baseDir}/query.py similar <id> --collection models [--limit 5]

Add or update content (upsert)

python3 {baseDir}/query.py add --collection models --id <id> --text "..." [--metadata '{"key": "value"}']

This command performs an upsert: if the ID already exists, it updates the content; otherwise, it creates a new entry.

List collections

python3 {baseDir}/query.py collections

Create a new collection

python3 {baseDir}/query.py create-collection <name>

Delete an item

python3 {baseDir}/query.py delete <id> --collection models

Collections

Pre-defined collections for different use cases:

Collection	Purpose
models	LoRA model descriptions (name, category, style notes)
memories	Bot memories and learned context
notes	User notes and reference material
posts	Posted content for semantic similarity checking

Environment Variables

All variables are optional with sensible defaults:

• CHROMADB_HOST: ChromaDB server hostname (default: localhost)
• CHROMADB_PORT: ChromaDB server port (default: 8000)
• CHROMADB_PATH: Path for persistent storage (optional, uses server if not set)

Default Mode: The skill connects to the ChromaDB server at localhost:8000 by default. To use persistent local storage instead, set CHROMADB_PATH=/path/to/storage.

Metadata

Each item can have metadata for filtering:

python3 {baseDir}/query.py add \
    --collection models \
    --id creature_v2 \
    --text "A LoRA for generating fantasy creatures with detailed scales and claws" \
    --metadata '{"category": "creatures", "base_model": "flux", "quality": 5}'

Query with metadata filter:

python3 {baseDir}/query.py query "dragon" --collection models --where '{"category": "creatures"}'

Examples

# Find models similar to a concept
python3 {baseDir}/query.py query "cyberpunk android with neon lights" --collection models

# Find posts similar to a new draft (avoid duplicates)
python3 {baseDir}/query.py query "Check out this new model release!" --collection posts

# Store a memory
python3 {baseDir}/query.py add \
    --collection memories \
    --id "mem_20250207_001" \
    --text "User prefers dark fantasy aesthetics with blue and purple color schemes"

# Find related memories
python3 {baseDir}/query.py query "color preferences" --collection memories

Server vs Persistent Mode

The skill supports two modes:

1. Server mode (default): Connects to a ChromaDB server at localhost:8000

   • No configuration needed if server is running locally
   • Set CHROMADB_HOST and CHROMADB_PORT for remote servers
   • Better for multi-agent access and concurrent operations

2. Persistent mode: Uses local file storage

   • Set CHROMADB_PATH=/path/to/storage to enable
   • Suitable for offline or single-agent setups

Troubleshooting

Connection refused

ChromaDB server not running at the configured host:port. Start it or check CHROMADB_HOST/CHROMADB_PORT.

Collection not found

The collection may not exist yet. Use python3 {baseDir}/query.py create-collection <name> to create it.

Empty results

• Lower your search threshold — semantic search can be sensitive to phrasing
• Try different query terms or shorter queries
• Verify the collection has data: python3 {baseDir}/query.py collections