Discord bot with a multi-LLM backend and restricted jailbreak personas

What it does

This Python bot is essentially a switchboard between Discord and several large language model APIs. Users can chat, generate images with /draw, and swap backends—OpenAI, Claude, Gemini, Grok, or a free provider—from inside the server. It also supports configurable personas, including three admin-locked jailbreak modes that bypass standard safety filters.

The interesting bit

The project recently ditched cookie-based authentication entirely, calling it unreliable, and now treats the free provider as an unstable fallback rather than a main event. That honesty about fragility is rarer than you’d think in wrapper projects, and the Docker hardening—non-root user, read-only filesystem—suggests the author actually expects this to run in semi-hostile environments.

Key highlights

• Switch between five AI providers using a /provider command.
• Image generation through OpenAI DALL-E 3 or Google Gemini, with free alternatives if premium providers are unavailable.
• Persona system ranges from standard tones to three admin-only jailbreak modes gated by Discord user ID.
• No cookie-based authentication; all premium access uses environment variables.
• Docker deployment runs as a non-root user with a read-only filesystem.

Caveats

• The free provider is explicitly labeled unstable and serves outdated models roughly comparable to older GPT-3.5 or GPT-4 versions.
• Jailbreak personas can generate content that bypasses normal AI safety measures, so the admin-gating is there for a reason.
• Image generation quality and availability depend on which provider is currently active.

Verdict

Worth a look if you moderate a Discord community and want one bot that can route to multiple LLMs or test cheap fallbacks. Skip it if you need deep custom logic beyond chat, image generation, and persona switching—this is glue code, albeit tidy glue.