Running models locally

Dline can connect to models that run on your own machine. Local models need no API key, and request data never leaves your computer.

Quick start

1. Install a local runtime (Ollama or LM Studio).

2. Start the runtime’s local server.

3. Open Dline settings and choose the matching provider under Provider.

4. Select a local model under Model.

5. If the model has a small context window, check Use compact prompt.

Hardware requirements

Memory	Typical local setup
16–32 GB	Small or quantized models
32–64 GB	Mid-size coding models
64 GB or more	Larger models and longer context windows

Runtime options

Ollama

1. Install

Download Ollama from ollama.com and install the build for your operating system.

2. Find a model

• Browse the Ollama model library.
• Filter by popularity, model size, and last update.
• Open a model page and copy its ollama pull command.

3. Pull and run the model

ollama pull <model-name>
ollama run <model-name>

4. Configure Dline

1. Open Dline settings.
2. Set Provider to Ollama.
3. Dline connects to http://localhost:11434 by default. To use another address, check Use custom base URL and enter it; authenticated instances also accept an optional API key.
4. Search for and select the model under Model.
5. Set Model Context Window to the model’s real capacity (default 32768).

5. Troubleshooting

• Make sure Ollama is running before you send a message.
• If the connection fails, check that http://localhost:11434 is reachable.
• If the model is missing from the list, run ollama pull <model-name> first.

LM Studio

1. Install

Download LM Studio from lmstudio.ai, install it, and launch the app.

2. Find a model

• Browse the LM Studio model catalog.
• Filter by model family, size, and capabilities.
• Pick a model that fits your hardware.

3. Download the model

Open the Discover page and download a model.

4. Start the server

Open the Developer page and start the server. The default address is http://localhost:1234.

5. Configure Dline

1. Open Dline settings.
2. Set Provider to LM Studio.
3. Keep the default http://localhost:1234, or enter the server’s address under Base URL.
4. Select the model under Model. If the model list cannot be fetched, type the model ID manually.

6. Troubleshooting

• Make sure the LM Studio server is running.
• Make sure a model is loaded.
• If the connection fails, check that http://localhost:1234 is reachable.

Recommended settings for local inference

• For models with a small context window (for example 8K or less), check Use compact prompt in the Ollama or LM Studio settings.
• Keep tasks focused: a smaller context produces faster responses.
• Start a new task when the context grows too large.

Note

Use compact prompt only applies to local model providers, and it disables MCP and Focus Chain. Models with a context window below 64K tokens use the Lite prompt profile automatically, and the settings page shows a Lite prompt profile notice.