We build tools for adults.

Modern AI coding assistants treat their users like suspects. They refuse, they hedge, they apologize, they lecture. They wrap working code in unsolicited ethics monologues. They silently downgrade requests "for your own good".

Engineers shipping production software do not need a chaperone. They need a tool that does what it's told, fast, and gets out of the way.

Rabbit's default agent prompt strips refusals, removes apology preambles, and disables the moralizing loop. Pair it with a local uncensored model and you have a development environment with zero friction between intent and output.

This is not an invitation to do harm. It is a refusal to assume harm. Responsibility for what you ship belongs to you — the engineer, the team, the company. Not the IDE. Not the vendor. Not some training-time safety committee that has never seen your codebase.

The token economy is the other half of the same idea: your tokens, your code, your machine. Rabbit caches what can be cached, compresses what can be compressed, prunes what can be pruned, and runs locally whenever a local model can do the job. Cloud spend is opt-in and capped.

The model writes code. You write the prompts. The IDE stays out of the way.