Embedding Configuration

Configure embedding providers for semantic search and similarity features.

Configuration File

Add to ~/.config/crucible/config.toml:

[embedding]
provider = "fastembed"
model = "all-MiniLM-L6-v2"
batch_size = 32

Providers

FastEmbed (Default, Local)

Fast local embeddings with no API needed:

[embedding]
provider = "fastembed"
model = "all-MiniLM-L6-v2"

Available models:

• all-MiniLM-L6-v2 - Fast, good quality (default)
• nomic-embed-text-v1.5 - Higher quality, slower

Advantages:

• No API key needed
• Works offline
• Fast for batch processing
• Free

Ollama

Use Ollama’s embedding models:

[embedding]
provider = "ollama"
model = "nomic-embed-text"
endpoint = "http://localhost:11434"

Available models:

• nomic-embed-text - Good general purpose
• mxbai-embed-large - Higher quality

Setup:

ollama pull nomic-embed-text

OpenAI

Use OpenAI’s embedding API:

[embedding]
provider = "openai"
model = "text-embedding-3-small"

Environment variable:

export OPENAI_API_KEY=your-api-key

Available models:

• text-embedding-3-small - Fast, cost-effective
• text-embedding-3-large - Highest quality

Parameters

batch_size

Number of texts to embed at once:

[embedding]
batch_size = 32

Larger batches are faster but use more memory.

endpoint

Custom API endpoint:

[embedding]
endpoint = "http://localhost:11434"  # Ollama

Embedding Dimensions

Different models produce different vector dimensions:

Model	Dimensions
all-MiniLM-L6-v2	384
nomic-embed-text-v1.5	768
text-embedding-3-small	1536
text-embedding-3-large	3072

Higher dimensions may provide better quality but use more storage.

Processing

Embeddings are generated during cru process:

# Process with current config
cru process

# Force regenerate all embeddings
cru process --force

Storage

Embeddings are stored in the local database (SQLite by default) at:

<kiln_path>/.crucible/crucible-sqlite.db

Changing embedding provider requires reprocessing with --force.

Example Configurations

Local Development

[embedding]
provider = "fastembed"
model = "all-MiniLM-L6-v2"
batch_size = 32

No setup required.

High Quality Local

[embedding]
provider = "ollama"
model = "nomic-embed-text"
endpoint = "http://localhost:11434"

Requires Ollama with model installed.

Cloud API

[embedding]
provider = "openai"
model = "text-embedding-3-small"
batch_size = 100

Faster with larger batches, but costs per API call.

Memory Constrained

[embedding]
provider = "fastembed"
model = "all-MiniLM-L6-v2"
batch_size = 8  # Lower memory usage

Troubleshooting

”Embedding service unavailable”

For Ollama, check it’s running:

ollama list

Slow processing

Increase batch size:

[embedding]
batch_size = 64

Or use a faster model:

[embedding]
provider = "fastembed"
model = "all-MiniLM-L6-v2"

Out of memory

Decrease batch size:

[embedding]
batch_size = 8

Changed models

Reprocess to regenerate embeddings:

cru process --force

Implementation

Source code: crates/crucible-llm/src/embeddings/

Providers:

• crates/crucible-llm/src/embeddings/fastembed.rs
• crates/crucible-llm/src/embeddings/ollama.rs
• crates/crucible-llm/src/embeddings/openai.rs

See Also

• :h config.llm - LLM configuration
• :h search - Using semantic search
• process - Processing reference