Hello everyone,

If you want to use the ScriptExpert AI Agent without sending your code to an external provider or paying API fees, you can connect ScriptExpert to a local Ollama server.

This tutorial covers the complete setup, including hardware requirements, Ollama installation, downloading a coding model and connecting it to ScriptExpert.

Minimum hardware

For acceptable coding performance, we recommend:

• 6-core or better processor

• 32 GB RAM

• NVIDIA or AMD GPU with at least 8 GB VRAM

• NVMe SSD with at least 50 GB free space

• Windows 10/11 or a modern Linux distribution

Ollama can run without a dedicated GPU, but coding models will respond considerably more slowly.

Recommended model configurations:

• 16 GB RAM, CPU only: qwen2.5-coder:3b

• 8 GB VRAM and 32 GB RAM: qwen2.5-coder:7b

• 12–16 GB VRAM and 32–64 GB RAM: qwen2.5-coder:14b

• 24 GB VRAM and 64 GB RAM: qwen3-coder:30b

Installing Ollama on Linux

On Ubuntu, Debian or Linux Mint, open a terminal and run:

curl -fsSL https://ollama.com/install.sh | sh

Enable and start the service:

sudo systemctl enable ollama
sudo systemctl start ollama
sudo systemctl status ollama

Verify that the API is working:

curl http://127.0.0.1:11434/api/tags

Installing Ollama on Windows

Download Ollama from:

https://ollama.com/download/windows

Install and start the application. Ollama will run in the background and expose its local API at:

http://localhost:11434

Verify it from PowerShell:

ollama --version
Invoke-RestMethod "http://localhost:11434/api/tags"

Downloading a coding model

For most users, we recommend starting with:

ollama pull qwen2.5-coder:7b

For more powerful hardware:

ollama pull qwen2.5-coder:14b

Or, for a server with approximately 24 GB VRAM:

ollama pull qwen3-coder:30b

Test the model:

ollama run qwen2.5-coder:7b

Ask it to generate or explain some code. To close the interactive session, enter:

/bye

You can list all installed models with:

ollama list

Using Ollama on another computer

You can install Ollama on a dedicated Linux server and connect to it from a Windows or Linux computer running ScriptExpert.

On the Ollama server, run:

sudo systemctl edit ollama

Add:

[Service]
Environment="OLLAMA_HOST=0.0.0.0:11434"
Environment="OLLAMA_CONTEXT_LENGTH=16384"
Environment="OLLAMA_NUM_PARALLEL=1"
Environment="OLLAMA_MAX_LOADED_MODELS=1"

Restart Ollama:

sudo systemctl daemon-reload
sudo systemctl restart ollama

Allow access only from your local network:

sudo ufw allow from 192.168.1.0/24 to any port 11434 proto tcp

Replace the subnet with the one used by your network.

Important: Ollama’s local API does not require authentication. Never expose port 11434 directly to the public internet. For access outside your home or office, use a private VPN such as Tailscale or WireGuard.

Test the connection from the ScriptExpert computer:

curl http://192.168.1.100:11434/api/tags

Replace 192.168.1.100 with the IP address of your Ollama server.

Connecting Ollama to ScriptExpert

Open ScriptExpert Code Editor and navigate to the AI provider settings.

Use the following configuration when Ollama runs on the same computer:

Provider: Ollama
Ollama URL: http://localhost:11434
Plan model: qwen2.5-coder:7b
Act model: qwen2.5-coder:7b

For a separate Ollama server:

Ollama URL: http://192.168.1.100:11434

Do not add /v1 to the URL. ScriptExpert automatically uses the correct Ollama API endpoints.

Press the refresh button, select the installed model for Plan and Act, and save the settings.

On systems with limited VRAM, use the same model for both Plan and Act. Using two different large models can cause repeated model loading and slower responses.

Checking GPU acceleration

While the model is running, execute:

ollama ps

The PROCESSOR column should preferably show:

100% GPU

For NVIDIA GPUs, you can also monitor usage with:

nvidia-smi

If the model runs mainly on the CPU, use a smaller model or reduce the configured context length.

Troubleshooting

If ScriptExpert cannot connect:

sudo systemctl status ollama
sudo ss -ltnp | grep 11434
curl http://127.0.0.1:11434/api/tags

If no models appear in ScriptExpert:

ollama list
ollama pull qwen2.5-coder:7b

Then return to ScriptExpert and refresh the model list.

If Ollama runs out of memory, reduce the context length:

Environment="OLLAMA_CONTEXT_LENGTH=8192"

Then restart the Ollama service.

Final recommendation

A good starting configuration for ScriptExpert is:

32 GB RAM
8–12 GB GPU VRAM
qwen2.5-coder:7b
16K context
One parallel request
The same model for Plan and Act

This provides a good balance between performance, code quality and memory usage.

If you encounter connection or model-loading problems, reply to this topic with your operating system, GPU model, available RAM, selected Ollama model and the output of ollama ps.

Official resources:

• Ollama documentation: https://docs.ollama.com

• Ollama model library: https://ollama.com/library

• ScriptExpert: https://scriptexpert.net