Use CyberStrike with abliteration.ai

Use the CyberStrike offensive security agent with abliteration.ai — CyberStrike is a fork of OpenCode, so authenticate, pick Abliterated Model Large V2, and choose a reasoning level.

CyberStrike is a fork of OpenCode, so setup is identical — abliteration.ai is a built-in provider (via models.dev), and there’s no config to write: authenticate, pick the model, and choose a reasoning level. CyberStrike is an open-source AI agent for offensive security that runs in your terminal.

Install

curl -fsSL https://cyberstrike.io/install.sh | bash

Or with npm: npm install -g @cyberstrike-io/cyberstrike@latest.

Authenticate

Get an ak_... key from the console, then:

cyberstrike auth login

Select abliteration.ai from the provider list and paste your key. (You can also do this inside CyberStrike with the /connect command, or by setting the ABLIT_KEY environment variable.)

The base URL is filled in automatically from the provider entry — you don’t need to set it.

Choose the model

In CyberStrike:

1. Run /models.
2. Open the provider list (Ctrl+A by default — the current keybind is shown in the dialog footer) and choose abliteration.ai.
3. Choose Abliterated Model Large V2 (or Abliterated Model for the multimodal base model).

Set the reasoning level

Reasoning depth is a model variant. Cycle it with Ctrl+T, or run /variants to pick from a list:

• Abliterated Model Large V2 — none, low, high, max
• Abliterated Model Large — none, high, max
• Abliterated Model (base) — none, minimal, low, medium, high, xhigh, max

The selection maps to the reasoning_effort request field. See thinking & reasoning effort for what each level does.

For Large V2, the none variant runs at low reasoning and hides the trace because the model cannot disable reasoning.

abliterated-model-large-v2 is text-only. Use abliterated-model if you need image inputs. If abliteration.ai doesn’t appear in the provider list yet, refresh the models cache with cyberstrike models --refresh.