Open Sandbox

Inspiration

My biggest inspiration comes from the open source community, I boom period of AI.Open source community work like it is on steriods,means tech giants release something and OS community design a alternative at no cost.But when you have a lot of data,you can't understand or use 1000s of great repository,even thinking about it impractical.So my real motivation comes from this great work.Why don't anyone can use this power all at one place and only work for innovation. This thing help in innovation and reduce carbon emission to create things from screatch because it will reduce the work time by 70% minimum in IT sector,wether for opensource community or startups which need to design a new thing based on existing features or projects.

What it does

OpenSandbox is a decentralized agentic low code platform,where user can use the potential from open source work or its own codebase and use its out of box innovative thinking to build something inceadible , for example fetch ai is a agentic work with lot of great features but not usefull for every person, like their is a person who is not great at coding and want to use fetch ai. So user face a challenge here, to overcome this problem, user can own low code platform which use fetch ai and another project named langflow which is also a agentic framework but built on langchain and drag and drop playground, so why not use drag and drop platground with fetch ai , so user can add drag and drop playground features and can built a new featured library with playground feature and fetch ai power just with some NLP commands,truth is that this example is like a needle in haystack.User have the whole community power in its hands only its thinking resist it,infact user can design something which we can't even imagine, just in minutes.Open sandbox is build from two parts one is analyser which analyse the repo by analyse its each and every file and directory without rely on documentation,their can be hundreds or thousands of repo can be used in a single project, second part is code genrator which understand it prebuild feature and redesign it according to need of user project without copy any code making it free from copyrights.Both parts are connected and controled by a parent model in a controlled sandboxed environment.

How we built it

We use the agents and llm like gemini ,chatgpt and other for best analysis of each and every repo, find their workflow,connections and compare the functionality of all repos under one parent controler llm, this including sujjestions for adding or improvement according to project.We design this in two parts one is responsible for repo analysis , repo can be 1 or hundred based on user, analyse each and every repo,their features,workflow etc.then based on user's need it findout the features from all repos,and find and store the file and directories where they are defined for future use,this part also give a chat option for proper and detailed analysis of any repo, which is based on repo summary and then by searching and analysing the required file and directories. Second part is generator part which is responsible for recreating,modifying features without copy code and any language according to user's need e.g. the feature is defined in python but user want to use that feature in a scale up project so modify that feature in java,etc.Due to non copy code method it is free from any copy right issues.This part generate,check and run that according to project and run the whole project for any type of errors or improvement. Then these both parts are connected and moniter by a llm which work as a overall coordinator.

Challenges we ran into

The biggest challenge we face is when their are a lot of repos,and files ,due to using free trail request per minute are not so high ,but when use premium account this will no longer be challenging. Another challenge is when files get too big, like million line code file then fit that into a single context window even in gemmini with 2M context window so use by spliting them into chunks which slow down the analysis process if their are too many large code files.

Accomplishments that we're proud of

This system is so much capable that it can modify itself too if allowed.

What we learned

The use of agents,llm,vector databases,etc.

What's next for Open Sandbox

This is a game changing project for IT sector when work on large code bases with millions lines of code in single file, to only analyse them it takes weeks.We'll modify our analysis llm with llama3 405B from groq for super speed analysis with 128k context window. And with some modification used for AI based search in various fields.