The Ollama plugin provides local model execution and can serve as a fallback option when cloud-based LLM providers are not configured. It requires running an Ollama server locally.

​
Features

  • Local execution - No API keys or internet required
  • Multiple models - Support for Llama, Mistral, Gemma, and more
  • Full model types - Text, embeddings, and objects
  • Cost-free - No API charges
  • Fallback option - Can serve as a local fallback when cloud providers are unavailable

​
Prerequisites

  1. Install Ollama
  2. Pull desired models: 
    ollama pull llama3.1
ollama pull nomic-embed-text

​
Installation

elizaos plugins add @elizaos/plugin-ollama

​
Configuration

​
Environment Variables

# Required
OLLAMA_API_ENDPOINT=http://localhost:11434/api

# Model configuration
# You can use any model available in your Ollama installation
OLLAMA_SMALL_MODEL=llama3.2                # Default: llama3.2
OLLAMA_MEDIUM_MODEL=llama3.1               # Default: llama3.1
OLLAMA_LARGE_MODEL=llama3.1:70b            # Default: llama3.1:70b
OLLAMA_EMBEDDING_MODEL=nomic-embed-text    # Default: nomic-embed-text

# Examples of other available models:
# OLLAMA_SMALL_MODEL=mistral:7b
# OLLAMA_MEDIUM_MODEL=mixtral:8x7b
# OLLAMA_LARGE_MODEL=llama3.3:70b
# OLLAMA_EMBEDDING_MODEL=mxbai-embed-large
# OLLAMA_EMBEDDING_MODEL=all-minilm

# Optional parameters
OLLAMA_TEMPERATURE=0.7

​
Character Configuration

{
  "name": "MyAgent",
  "plugins": ["@elizaos/plugin-ollama"]
}

​
Supported Operations

OperationModelsNotes
TEXT_GENERATIONllama3, mistral, gemmaVarious sizes available
EMBEDDINGnomic-embed-text, mxbai-embed-largeLocal embeddings
OBJECT_GENERATIONAll text modelsJSON generation

​
Model Configuration

The plugin uses three model tiers:
  • SMALL_MODEL: Quick responses, lower resource usage
  • MEDIUM_MODEL: Balanced performance
  • LARGE_MODEL: Best quality, highest resource needs
You can use any model from Ollama’s library:
  • Llama models (3, 3.1, 3.2, 3.3)
  • Mistral/Mixtral models
  • Gemma models
  • Phi models
  • Any custom models you’ve created
For embeddings, popular options include:
  • nomic-embed-text - Balanced performance
  • mxbai-embed-large - Higher quality
  • all-minilm - Lightweight option

​
Performance Tips

  1. GPU Acceleration - Dramatically improves speed
  2. Model Quantization - Use Q4/Q5 versions for better performance
  3. Context Length - Limit context for faster responses

​
Hardware Requirements

Model SizeRAM RequiredGPU Recommended
7B8GBOptional
13B16GBYes
70B64GB+Required

​
Common Issues

​
”Connection refused”

Ensure Ollama is running: 
ollama serve

​
Slow Performance

  • Use smaller models or quantized versions
  • Enable GPU acceleration
  • Reduce context length

​
External Resources