Python CLI for chatting with borrowed AI and fewer guardrails

What it does HacxGPT-CLI is a Python command-line chat client that aggregates multiple AI providers behind a single TUI. You supply your own keys for OpenRouter, Groq, or the project’s paid HacxGPT API, and the tool routes requests through a custom local api.py engine that deliberately avoids official SDKs like openai or litellm. It stores configuration in ~/.hacx, encrypts keys against a machine-specific hardware ID, and supports conversation save/load, reasoning-panel rendering for chain-of-thought tags, and in-chat auto-updates.

The interesting bit The project treats censorship as a configuration problem: it uses “advanced jailbreak prompts”—system-level prompt engineering—to try to make off-the-shelf models less trigger-happy with refusals. It also replaced standard API SDKs with a standalone local engine, though the README offers no technical detail on how that engine improves speed or control beyond the claim itself.

Key highlights

• Aggregates OpenRouter, Groq, and the HacxGPT production API in one interface.
• Custom local API engine with zero external SDK dependencies for provider calls.
• Machine-bound key encryption tied to hardware IDs; local-only storage in ~/.hacx.
• Cross-platform support including Linux, Windows, macOS, and Termux.
• Built-in session persistence with /save, /load, and /sessions commands.

Caveats

• This is strictly a wrapper framework, not a custom model; the README explicitly states all inference happens on third-party infrastructure.
• The “unrestricted” experience is entirely dependent on provider tolerance and prompt tricks, not on architecture changes to the models themselves.
• The README is vague on how the custom engine or encryption schemes work beyond high-level descriptions, and the marketing tone sometimes oversells the differentiation.

Verdict Try it if you want a terminal-native chat client that lets you hop between providers and experiment with refusal-bypass prompting. Pass if you need a local offline model or a fully transparent, auditable inference stack—this is a remote-API client with attitude.