starlet/ollama

Ollama provider for starlet.

Ollama is a local LLM runtime that supports many open-source models.

Usage

import starlet
import starlet/ollama

let client = ollama.new("http://localhost:11434")

starlet.chat(client, "qwen3:0.6b")
|> starlet.user("Hello!")
|> starlet.send()

Thinking Mode

For thinking-capable models (DeepSeek-R1, Qwen3), configure thinking:

starlet.chat(client, "deepseek-r1")
|> ollama.with_thinking(ollama.ThinkingEnabled)
|> starlet.user("Solve this step by step...")
|> starlet.send()

Types

Model

</>

Information about an available model.

pub type Model {
  Model(name: String, size: String)
}

Constructors

```
Model(name: String, size: String)
```

Thinking

</>

Thinking mode configuration for Ollama.

pub type Thinking {
  ThinkingEnabled
  ThinkingDisabled
  ThinkingLow
  ThinkingMedium
  ThinkingHigh
}

Constructors

```
ThinkingEnabled
```
Enable thinking (boolean mode)
```
ThinkingDisabled
```
Disable thinking
```
ThinkingLow
```
Low reasoning effort (for models that support effort levels)
```
ThinkingMedium
```
Medium reasoning effort
```
ThinkingHigh
```
High reasoning effort

Values

list_models

</>

pub fn list_models(
  base_url: String,
) -> Result(List(Model), starlet.StarletError)

Lists available models from the Ollama server.

new

</>

pub fn new(base_url: String) -> starlet.Client(@internal Ext)

Creates a new Ollama client with the given base URL.

Example: ollama.new("http://localhost:11434")

thinking

</>

pub fn thinking(
  turn: starlet.Turn(t, f, @internal Ext),
) -> option.Option(String)

Get the thinking content from an Ollama turn (if present).

with_thinking

</>

pub fn with_thinking(
  chat: starlet.Chat(t, f, s, @internal Ext),
  mode: Thinking,
) -> starlet.Chat(t, f, s, @internal Ext)

Configure thinking mode for thinking-capable models. When not set, the provider’s default applies (enabled for thinking models).