> ## Documentation Index
> Fetch the complete documentation index at: https://mintlify.com/pensarai/apex/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Providers

> Configure Anthropic, OpenAI, AWS Bedrock, OpenRouter, and local vLLM models for Pensar Apex

Pensar Apex supports multiple AI providers. **Anthropic models provide the best performance** and are recommended for optimal results.

## Supported Providers

Pensar Apex integrates with the following AI providers:

<CardGroup cols={2}>
  <Card title="Anthropic" icon="a" href="#anthropic-claude">
    Recommended provider with best pentesting performance
  </Card>

  <Card title="OpenAI" icon="robot" href="#openai">
    GPT-4 and other OpenAI models
  </Card>

  <Card title="AWS Bedrock" icon="aws" href="#aws-bedrock">
    Enterprise-grade AI on AWS infrastructure
  </Card>

  <Card title="OpenRouter" icon="route" href="#openrouter">
    Unified API for multiple model providers
  </Card>

  <Card title="vLLM (Local)" icon="server" href="#vllm-local-models">
    Self-hosted open-source models
  </Card>
</CardGroup>

## Configuration Methods

You can configure AI providers in two ways:

1. **Environment Variables** (recommended for CLI usage)
2. **TUI Configuration** (interactive setup)

<Note>
  Environment variables take precedence over TUI configuration. This allows you to override settings per-session.
</Note>

## Anthropic (Claude)

**Recommended** for best penetration testing performance.

### Setup

<Steps>
  <Step title="Get API Key">
    Sign up at [console.anthropic.com](https://console.anthropic.com) and create an API key.
  </Step>

  <Step title="Set Environment Variable">
    ```bash theme={null}
    export ANTHROPIC_API_KEY="sk-ant-..."
    ```
  </Step>

  <Step title="Verify Configuration">
    ```bash theme={null}
    pensar doctor
    ```

    You should see:

    ```
    ✓ Anthropic API key configured
    ```
  </Step>
</Steps>

### Recommended Models

* `claude-sonnet-4-5` (default) - Best balance of performance and cost
* `claude-opus-4` - Maximum reasoning capability for complex targets
* `claude-sonnet-3-5` - Previous generation, still highly capable

<Tip>
  For most pentesting scenarios, `claude-sonnet-4-5` provides excellent results at reasonable cost.
</Tip>

## OpenAI

Supports GPT-4 and other OpenAI models.

### Setup

<Steps>
  <Step title="Get API Key">
    Create an API key at [platform.openai.com/api-keys](https://platform.openai.com/api-keys).
  </Step>

  <Step title="Set Environment Variable">
    ```bash theme={null}
    export OPENAI_API_KEY="sk-..."
    ```
  </Step>

  <Step title="Select Model">
    When launching pentest, specify the model:

    ```bash theme={null}
    pensar pentest --target https://example.com --model gpt-4
    ```
  </Step>
</Steps>

### Available Models

* `gpt-4` - Most capable GPT model
* `gpt-4-turbo` - Faster, cost-effective alternative
* `gpt-3.5-turbo` - Budget option (not recommended for complex pentesting)

<Warning>
  GPT models may not perform as well as Claude for security testing tasks. Use Anthropic when possible.
</Warning>

## AWS Bedrock

Run models on AWS infrastructure with enterprise-grade security and compliance.

### Authentication Methods

Bedrock supports two authentication modes:

<Tabs>
  <Tab title="Bearer Token">
    Simple token-based auth:

    ```bash theme={null}
    export BEDROCK_API_KEY="your-bearer-token"
    export AWS_REGION="us-east-1"
    ```
  </Tab>

  <Tab title="IAM Credentials">
    Use IAM credentials (recommended for production):

    ```bash theme={null}
    export AWS_ACCESS_KEY_ID="AKIA..."
    export AWS_SECRET_ACCESS_KEY="..."
    export AWS_SESSION_TOKEN="..."  # optional, for temporary credentials
    export AWS_REGION="us-east-1"
    ```
  </Tab>
</Tabs>

### Supported Models

Bedrock provides access to Claude, Llama, and other models:

```bash theme={null}
# Use Claude on Bedrock
pensar pentest --target https://example.com --model anthropic.claude-sonnet-4-5

# Use Llama on Bedrock
pensar pentest --target https://example.com --model meta.llama3-70b
```

<Note>
  Model IDs on Bedrock may differ from direct provider APIs. Check the [AWS Bedrock documentation](https://docs.aws.amazon.com/bedrock/) for available models in your region.
</Note>

## OpenRouter

Unified API for accessing models from multiple providers (Anthropic, OpenAI, Google, Meta, etc.).

### Setup

<Steps>
  <Step title="Get API Key">
    Sign up at [openrouter.ai](https://openrouter.ai) and create an API key.
  </Step>

  <Step title="Set Environment Variable">
    ```bash theme={null}
    export OPENROUTER_API_KEY="sk-or-..."
    ```
  </Step>

  <Step title="Select Model">
    ```bash theme={null}
    pensar pentest --target https://example.com --model anthropic/claude-sonnet-4-5
    ```
  </Step>
</Steps>

### Model Format

OpenRouter uses `provider/model-id` format:

* `anthropic/claude-sonnet-4-5`
* `openai/gpt-4`
* `google/gemini-pro`
* `meta-llama/llama-3-70b`

See [openrouter.ai/models](https://openrouter.ai/models) for all available models.

## vLLM (Local Models)

Run open-source models locally with vLLM for complete data privacy.

### Setup

<Steps>
  <Step title="Start vLLM Server">
    ```bash theme={null}
    # Install vLLM
    pip install vllm

    # Start server with a model
    vllm serve meta-llama/Llama-3.1-70B-Instruct --port 8000
    ```
  </Step>

  <Step title="Configure Pensar Apex">
    Set the local model endpoint:

    ```bash theme={null}
    export LOCAL_MODEL_URL="http://localhost:8000/v1"
    ```
  </Step>

  <Step title="Specify Model Name">
    In the TUI, go to Models screen and enter the model name in "Custom local model (vLLM)" input.

    Or via CLI:

    ```bash theme={null}
    pensar pentest --target https://example.com --model meta-llama/Llama-3.1-70B-Instruct
    ```
  </Step>
</Steps>

<Warning>
  Local models may not perform as well as Claude for pentesting. Use for offline scenarios or when data privacy is critical.
</Warning>

See the [vLLM Setup Guide](/guides/vllm-setup) for detailed instructions.

## Provider Priority

When multiple providers are configured, Pensar Apex checks for API keys in this order:

1. `ANTHROPIC_API_KEY` (Anthropic)
2. `OPENAI_API_KEY` (OpenAI)
3. `OPENROUTER_API_KEY` (OpenRouter)
4. `BEDROCK_API_KEY` or AWS credentials (Bedrock)
5. `LOCAL_MODEL_URL` (vLLM)

<Tip>
  You can override the default by explicitly specifying `--model` on the command line.
</Tip>

## Configuration File

AI provider settings are stored in `~/.pensar/config.json`:

```json theme={null}
{
  "anthropicAPIKey": "sk-ant-...",
  "openAiAPIKey": null,
  "openRouterAPIKey": null,
  "bedrockAPIKey": null,
  "localModelUrl": null,
  "localModelName": null,
  "selectedModelId": "claude-sonnet-4-5",
  "responsibleUseAccepted": true
}
```

<Warning>
  Environment variables always take precedence over config file values.
</Warning>

## Troubleshooting

<AccordionGroup>
  <Accordion title="'No AI provider configured' error">
    Ensure at least one AI provider API key is set:

    ```bash theme={null}
    # Check current config
    pensar doctor

    # Set API key
    export ANTHROPIC_API_KEY="your-key"
    ```
  </Accordion>

  <Accordion title="Rate limit errors">
    Pensar Apex automatically retries on rate limits. If you hit persistent rate limits:

    * **Anthropic**: Upgrade your plan tier at [console.anthropic.com](https://console.anthropic.com)
    * **OpenAI**: Check usage limits at [platform.openai.com/account/limits](https://platform.openai.com/account/limits)
    * **Bedrock**: Request quota increases in AWS Service Quotas
  </Accordion>

  <Accordion title="vLLM connection failed">
    Verify the vLLM server is running:

    ```bash theme={null}
    curl http://localhost:8000/v1/models
    ```

    If this fails, ensure vLLM is started and accessible.
  </Accordion>

  <Accordion title="AWS Bedrock authentication failed">
    For IAM credentials, verify:

    ```bash theme={null}
    aws sts get-caller-identity
    ```

    Ensure your IAM user/role has `bedrock:InvokeModel` permission.
  </Accordion>
</AccordionGroup>

## Next Steps

<CardGroup cols={2}>
  <Card title="Model Selection" icon="sliders" href="/configuration/models">
    Choose the right model for your testing needs
  </Card>

  <Card title="Environment Variables" icon="terminal" href="/configuration/environment-variables">
    Complete reference of all configuration options
  </Card>

  <Card title="vLLM Setup" icon="server" href="/guides/vllm-setup">
    Detailed guide to self-hosting models
  </Card>

  <Card title="Run First Pentest" icon="rocket" href="/quickstart">
    Start testing with your configured provider
  </Card>
</CardGroup>