Manhattan Kernel

A deterministic Rust kernel that isolates dangerous LLM agents inside a strict compiler-integrated safety sandbox.

Project Details

Manhattan Kernel is a small solo research project. It is a deterministic runtime that works together with a language model. The model can give suggestions, but it does not run code or decide if its own results are correct. Every important change is checked by the compiler and by the structure system before it can be used. I strted this project because AI agents are becoming more capable and they get more access to tools and files. I think we need more reliable ways to check what they do instead of simply trusting every output. At the moment there is a working Rust prototype with a task system, Kernel Structure Graph (KSG), deterministic fingerprints and a system that tries local solutionsz before asking the language model. With this grant I want to finish the schema composition part, add more repair methods, connect a real language model and compare the results with normal LLM-based workflows. I also want to test it on CPU-only hardware to keep it simple and accessible. I work on this project alone. There is no team, company or co-founder behind it. I use AI mainly for coding help and sometimes because English is not my first language, but the technical decisions and direction are made by me. My background is in security work, and I have been seriously learning software development since 2020. The result will be a working schema search engine, performance tests, and public documentation where others can see how it works and try it themselves.

https://github.com/lipcsiklaszlo5-lgtm/DrManhattan-Project

Theory of Impact

Manhattan Kernel follows one simple idea, the model can suggest things, but it should not decide alone if something is correct. Every result is checked with fixed methods before it is used.

People

László Lipcsik

Team Member

Grants Received– no grants recorded

Add grant

Funding Asks

grantmaking.ai Launch Round

Applied

Minimum

$10,000

Ideal

$20,000

How the money will be spent

The minimum amount of $10,000 would give me enough time to work on Manhattan Kernel and improve the current prototype. The ideal amount of $20,000 would let me work on it longer, do more testing and prepare a better public version. The money would be used mainly for my development time, hardware and testing costs, documentation, and small technical expenses. Around $7,000–15,000 would support the time I spend on planning, coding, testing and improving the system. Since this is a solo project, the biggest cost is having enough time to focus on development. Around $1,000–2,000 would be used for hardware and testing. The project is designed to run on CPU-based machines, so I want to test it on different setups and improve reliability.

Around $1,000–2,000 would be used for documentation, examples and preparing the project for public use.

Around $500–1,000 would cover small technical costs like basic services and hosting. With $10,000 I can complete the main development goals, improve the prototype and run the first serious tests. With $20,000 I can spend more time improving the system, doing more testing and preparing a more complete public release. The funding is not for building a company or hiring a team. It is mainly to support my own work and turn the current prototype into a finished open research project.

Discussion

No comments yet. Be the first to share your thoughts.