> ## Documentation Index
> Fetch the complete documentation index at: https://docs.profclaw.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Ollama

> Run local AI models with Ollama. Zero API costs, full privacy, works offline. Default provider when no cloud keys are set.

Ollama lets you run open-source models locally. profClaw connects to Ollama automatically - it's always enabled and is the fallback when no cloud API keys are configured.

## Supported Models

Ollama supports hundreds of models. Popular choices in profClaw:

| Alias             | Model          | Best For        |
| ----------------- | -------------- | --------------- |
| `local` / `llama` | llama3.2       | General purpose |
| `deepseek-local`  | deepseek-r1:7b | Reasoning tasks |
| `qwen`            | qwen2.5:14b    | Multilingual    |
| `mistral-local`   | mistral:7b     | Fast inference  |

Any model available in `ollama list` can be used by its full name.

<Note>
  Most local Ollama models do not support native tool calling. profClaw automatically falls back to manual tool prompting for these models.
</Note>

## Setup

<Steps>
  <Step title="Install Ollama">
    ```bash theme={null}
    # macOS
    brew install ollama

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

    # Windows: download from https://ollama.com
    ```
  </Step>

  <Step title="Pull a model">
    ```bash theme={null}
    ollama pull llama3.2
    ollama pull deepseek-r1:7b
    ollama pull qwen2.5:14b
    ```
  </Step>

  <Step title="Start the Ollama server">
    ```bash theme={null}
    ollama serve
    # Runs at http://localhost:11434 by default
    ```
  </Step>

  <Step title="profClaw connects automatically">
    No API key needed. profClaw connects to Ollama at `http://localhost:11434`.

    ```bash theme={null}
    profclaw doctor --provider ollama
    ```
  </Step>
</Steps>

## Environment Variables

<ParamField path="OLLAMA_BASE_URL" type="string">
  Ollama server URL. Defaults to `http://localhost:11434`. Override for remote Ollama instances.
</ParamField>

## Configuration Example

<Tabs>
  <Tab title=".env (remote Ollama)">
    ```bash theme={null}
    OLLAMA_BASE_URL=http://192.168.1.100:11434
    ```
  </Tab>

  <Tab title="settings.yml">
    ```yaml theme={null}
    providers:
      ollama:
        base_url: "${OLLAMA_BASE_URL}"
        default_model: "llama3.2"
    ```
  </Tab>

  <Tab title="Docker Compose">
    ```yaml theme={null}
    services:
      profclaw:
        image: profclaw/profclaw:latest
        environment:
          OLLAMA_BASE_URL: "http://ollama:11434"
      ollama:
        image: ollama/ollama:latest
        volumes:
          - ollama_data:/root/.ollama
    volumes:
      ollama_data:
    ```
  </Tab>
</Tabs>

## Model Aliases

| Alias            | Model          |
| ---------------- | -------------- |
| `local`          | llama3.2       |
| `llama`          | llama3.2       |
| `deepseek-local` | deepseek-r1:7b |
| `qwen`           | qwen2.5:14b    |
| `mistral-local`  | mistral:7b     |

## Usage Examples

<CodeGroup>
  ```bash CLI theme={null}
  # Use default local model
  profclaw chat --model local "What is dependency injection?"

  # Use a specific model by name
  profclaw chat --model llama3.2 "Explain this code"

  # Use any installed model
  profclaw chat --model codellama:13b "Write a binary search"
  ```

  ```typescript SDK theme={null}
  import { chat } from 'profclaw';

  const response = await chat("Summarize this text", {
    model: "local",
  });
  ```
</CodeGroup>

## Notes

* Ollama is always lowest priority in auto-selection. If any cloud key is set, it takes precedence.
* Local models work without internet access - useful for air-gapped environments.
* GPU acceleration significantly improves performance. Ollama auto-detects CUDA/Metal.
* Tool calling is available via manual prompting fallback for models that don't support it natively.

## Related

* [AI Providers Overview](/ai-providers/overview) - Compare all 37 supported providers
* [LM Studio](/ai-providers/lmstudio) - Alternative local model runner with a GUI
* [Local LLM Guide](/guides/local-llm) - Run profClaw with fully local models
* [profclaw provider](/cli/provider) - Add and test providers from the CLI