AstraCode plans a change across your repository, writes it, runs your tests and hands you a diff to approve. Every model - Anthropic, OpenAI, xAI - sits in one picker, routed per task against one allowance.
The part people notice later is the meter. It prices each step at the rate that actually applied, including long-context repricing, where a model charges the higher rate on the ENTIRE request past a token line. One agent run clears it. Most meters miss it.
Free tier, no card.
