If you keep 1 Claude or any other agentic coding session too long, it starts halucinating and starts making more and more mistakes If you start a new chat, it will spend lots of tokens trying to read files and folder structure trying to regain context. Even then, with only the parts it had read will make mistakes. My tool solves the problem, My tool summarizes the code into a 3d json context file. It has the complete code-base, with descriptions, and dependencins, The AI can just look at this 1 file and it will know what the project is about, how it is structured, how to modify it, and where will I find the code that I need to change and also most importantly, if I change it,, what other files might get effected because they are using it too It will then proceed to read the exact file it needs to read and impliment the changes, without blidly looking here and there trying to understand the structure and wasting tokens in output thinking and input file reading tokens Try it out using npm i -g reporose cd project/root reporose analyze you may optionally do reporose serve to view the map but this is just for you and not the AI for the descriptions you can use Ollama cloud, go to https://ollama.com/ and get a free api key it will use it to generate small descriptions of the file so claude knows exactly what file does what It also supports local ai, claude and open ai api, groq api and no ai descriptions mode pssst Show love by giving me a star on my repo https://github.com/MIbrahimPro/R... Thank you so much!!!

RepoRose

Start new claude chat with full context without using tokens

Start new claude chat with full context without using tokens

Forum Threads

Suggestions and Issues

Forum Threads

Suggestions and Issues