Quick hack to save 90% on LLM tokens when parsing iOS builds
Hey guys,
If you’ve ever tried using LLMs to audit or check compliance on iOS .ipa builds, you know it absolutely destroys your token limits. A standard RAG setup usually eats up around 37k tokens just to parse the boilerplate and structure.
While building ipaShip, we managed to cut that down to just 2,500 tokens (about 7% of the cost). Instead of dumping the whole file extraction into the prompt, the system strips it down to a "structural skeleton" first, then feeds the LLM only the exact policy context it needs.
It supports OpenAI, Anthropic, Gemini, and OpenRouter using your own API keys, and the files self-destruct after the check.
If you want to run quick, cheap pre-submission checks on your apps, give it a spin at ipaShip.com. Let me know what you think!
Product Link - https://opensource.gracias.sh/
GitHub - https://github.com/atharvnaik1/ipaship-audit