​
Introduction

Ollama allows you to run large language models locally on your own hardware. By integrating Ollama with Aikeedo, you can provide AI capabilities without relying on cloud services, ensuring data privacy and reducing costs.
Currently, Ollama integration is available only for the Chat tool in Aikeedo.

​
About Ollama

Ollama is an open-source project that simplifies running and managing large language models locally. It offers:
  • Easy installation and setup
  • Support for various open-source models
  • Local execution for enhanced privacy
  • Custom model configuration
  • REST API for integration
For more information, visit the official Ollama website.

​
Setting Up Ollama

​
Step 1: Install Ollama

curl -fsSL https://ollama.ai/install.sh | sh
For Apple Silicon (M1/M2) Macs, Metal GPU acceleration is supported automatically.
For detailed installation instructions and troubleshooting, refer to the Ollama Installation Guide.

​
Step 2: Configure Your Model

  1. Start the Ollama service
  2. Pull your desired model. For example:
ollama pull llama2
# or
ollama pull mistral
# or
ollama pull llama2-uncensored
View all available models at Ollama Model Library.

​
Step 3: Configure Server Address

By default, Ollama runs on: 
http://localhost:11434
For remote access, you’ll need to:
  1. Configure your firewall to allow access to port 11434
  2. Set up proper security measures as Ollama doesn’t include authentication by default
When exposing Ollama to remote access, ensure you implement appropriate security measures to protect your server.

​
Integrating with Aikeedo

​
Step 1: Configure Aikeedo

  1. Log in to your Aikeedo admin panel
  2. Navigate to Settings → Integrations → Ollama
  3. Enter your Ollama server address (e.g., http://localhost:11434)
  4. Add your models:
    • Key: A unique identifier (e.g., ollama/llama2:latest)
    • Name: Display name for the model
    • Provider: Set as “Meta” for Llama models, or appropriate provider
  5. Click “Save changes”
The server address must include the scheme (http/https), host, and port number.

​
Available Models

Popular models you can use with Ollama include:
  • Llama 2: Meta’s open-source model series
  • Mistral: High-performance 7B model
  • Phi: Microsoft’s compact yet powerful model
  • Neural Chat: Optimized for conversation
  • Code Llama: Specialized for programming tasks
  • Vicuna: Fine-tuned for chat and assistance
For a complete list, check the Ollama Model Library.

​
Model Configuration

You can customize model behavior using Modelfiles. Example: 
FROM llama2
PARAMETER temperature 0.7
PARAMETER top_p 0.9
PARAMETER top_k 40
Learn more about model configuration in the Ollama Documentation.

​
Hardware Requirements

Minimum requirements vary by model:
  • 7B models: 8GB RAM
  • 13B models: 16GB RAM
  • 33B+ models: 32GB+ RAM
GPU acceleration is supported for:
  • NVIDIA GPUs with CUDA
  • AMD GPUs with ROCm
  • Apple Silicon (Metal)

​
Best Practices

  1. Model Selection:
    • Start with smaller models (7B) and test performance
    • Consider your hardware capabilities
    • Choose models based on your specific use case
  2. Performance:
    • Enable GPU acceleration when available
    • Monitor system resources
    • Adjust model parameters for optimal performance
  3. Security:
    • Implement proper firewall rules
    • Use reverse proxy for additional security
    • Regular system updates

​
Troubleshooting

Common issues and solutions:
  1. Connection Issues:
    • Verify Ollama service is running
    • Check server address format
    • Confirm firewall settings
  2. Model Loading Failures:
    • Ensure sufficient system resources
    • Verify model is properly pulled
    • Check model compatibility
  3. Performance Issues:
    • Monitor system resources
    • Consider using a smaller model
    • Adjust model parameters

​
Additional Resources

Keep Ollama and your models updated for the best performance and security.