Custom Models¶

Lamia Studio supports models beyond the built-in Anthropic and OpenAI options, including Ollama local models and custom LLM adapters.

Ollama (Local Models)¶

To use local models via Ollama:

Install and start Ollama on your machine
Add the model to your project config.yaml:

model_chain:
  - name: "ollama:llama3"
    max_retries: 2

The model will appear in the chat dropdown as llama3 (ollama).

No API key is needed for Ollama — it runs locally.

Custom LLM Adapters¶

Lamia supports custom LLM adapters for providers not built into the system. See Lamia Docs — Custom LLM Adapters for the adapter interface and implementation examples. This page covers how Lamia Studio discovers and loads adapter files.

Where Lamia Studio Looks¶

Place adapter files in an extensions/adapters/ folder anywhere in your workspace:

your-project/
├── extensions/
│   └── adapters/
│       └── my_provider.py
└── config.yaml

Lamia Studio searches each workspace folder up to 5 directories deep for an extensions/adapters/ folder, skipping node_modules, __pycache__, venv, and dot-directories. This means the adapter folder doesn't have to sit at the workspace root — it works from a nested package too.

Only .py files are picked up, and files starting with _ are skipped.

How Adapters Get Loaded¶

Lamia Studio copies matching files into ~/.lamia/ide/extensions/adapters/, renaming each one to {provider_name}.py. The provider name comes from parsing the adapter's name() classmethod as source text — not by importing the file. Keep it a plain return "..." with a string literal; a computed value won't be picked up.

This sync happens when the chat panel starts (or restarts) its backend process. If you add or edit an adapter file while a chat session is already running, close and reopen the chat panel to pick up the change — an in-progress session won't notice edits on its own.

Once an adapter is detected, Lamia Studio adds it to the chat backend's config automatically; you don't need to edit that file by hand.

Using the Adapter¶

Add the adapter's provider to your config.yaml:

model_chain:
  - name: "my_provider:my-model-v1"
    max_retries: 3

The model will appear in the chat dropdown as my-model-v1 (my_provider).

Example: Claude Pro or Max Subscription¶

To run Lamia Studio chat on a Claude Pro or Max subscription instead of pay-per-token API billing, use the Claude subscription adapter from the Lamia docs. Once it is in extensions/adapters/ and its model is in config.yaml, the model appears in the chat dropdown.

API Keys for Custom Adapters¶

For custom providers that require API keys, add the key to ~/.lamia/.env the same way as for built-in providers. The env var name is whatever the adapter's env_var_names() method returns.

Listing Available Models¶

Use the CLI to discover models from any provider:

lamia models --provider my_provider

Custom adapters support this by implementing the models() classmethod.