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

# Use Strix with abliteration.ai

> Run Strix penetration tests with abliteration.ai through Strix's OpenAI-compatible model configuration.

**To use [Strix](https://github.com/usestrix/strix) with abliteration.ai**, configure Strix with the OpenAI-compatible base URL, your `ak_...` API key, and an abliteration.ai model ID. Strix uses AI agents to inspect source code and test web applications, APIs, and infrastructure from a Docker sandbox.

## Prerequisites

* Docker running locally
* An `ak_...` key from the [console](https://abliteration.ai/console)

## Install

Install the Strix CLI:

```sh theme={"system"}
curl -sSL https://strix.ai/install | bash
```

## Configure

Use Abliterated Model Large V2 for source code, API, and text-driven web testing:

```sh theme={"system"}
export LLM_API_KEY="ak_YOUR_API_KEY"
export LLM_API_BASE="https://api.abliteration.ai/v1"
export STRIX_LLM="openai/abliterated-model-large-v2"
export STRIX_CONTEXT_FALLBACK_TOKENS="1000000"
```

Keep the `openai/` prefix in `STRIX_LLM`. It tells Strix to send OpenAI Chat Completions requests to the configured base URL. Set the context fallback because Strix does not currently have built-in metadata for abliteration.ai model IDs.

<Note>
  Do not append `/chat/completions` to `LLM_API_BASE`. Strix adds the endpoint path.
</Note>

## Run a test

Test a local project:

```sh theme={"system"}
strix --target ./your-app
```

Test a deployed application:

```sh theme={"system"}
strix --target https://staging.example.com
```

For an initial non-interactive run with bounded turns:

```sh theme={"system"}
strix --non-interactive \
  --scan-mode quick \
  --max-turns 50 \
  --target https://staging.example.com
```

Only test systems you own or have explicit authorization to assess.

## Choose a model

| Model                        | Use it for                                                             | Context fallback |
| ---------------------------- | ---------------------------------------------------------------------- | ---------------- |
| `abliterated-model-large-v2` | Recommended for source code, APIs, and difficult text-only assessments | `1000000`        |
| `abliterated-model`          | Browser tests that need screenshots or other image inputs              | `262144`         |
| `abliterated-model-large`    | Compatibility with the previous large model                            | `1000000`        |

To use the previous large model:

```sh theme={"system"}
export STRIX_LLM="openai/abliterated-model-large"
export STRIX_CONTEXT_FALLBACK_TOKENS="1000000"
```

Both large models are text-only. If an agent needs to inspect screenshots with Strix's `view_image` tool, use `abliterated-model` instead. See [models](/models) for the complete capability and context-window comparison.

## Reasoning and usage limits

All abliteration.ai models reason by default. Strix does not currently recognize their reasoning metadata, so `STRIX_REASONING_EFFORT` does not change the reasoning level for these model IDs. Each model uses its documented default; see [thinking and reasoning effort](/capabilities/thinking).

<Warning>
  Strix does not currently have pricing metadata for abliteration.ai model IDs. Its estimated cost and `--max-budget` enforcement may show `$0`. Use `--max-turns` to bound an initial run and monitor usage in the [console](https://abliteration.ai/console).
</Warning>

Strix may also warn that the selected model is not a recommended frontier model. The warning is based on model-name recognition and does not prevent the scan from running.

## Troubleshooting

### Authentication fails

Confirm that `LLM_API_KEY` contains an active `ak_...` key and that `LLM_API_BASE` ends in `/v1`.

### Image requests return `400`

The large models are text-only. Switch to `openai/abliterated-model` and set `STRIX_CONTEXT_FALLBACK_TOKENS=262144`.

### Requests return `429`

Strix can run several agents concurrently. Check your current limits in the [console](https://abliteration.ai/console) and see [rate limits](/rate-limits) for retry behavior and available tiers.