I opened an old project and had no idea how I built it.
So I built Repo Recall: paste a GitHub URL, pick files, chat with them in the browser.
Stack: React, Express, PostgreSQL + pgvector, Gemini.
No clone. No editor. RAG does the rest.
Live: https://t.co/bcZ4GfPoXA
Code + write-up in the thread ๐งต
stop paying for courses in 2026
don't let these course sellers fool you, they will always be ready with another course, another mega sale, every year
these are not going to help you. trust me. and save yourself some money and work hard, use AI, you have all the knowledge you need out here
I opened an old project and had no idea how I built it.
So I built Repo Recall: paste a GitHub URL, pick files, chat with them in the browser.
Stack: React, Express, PostgreSQL + pgvector, Gemini.
No clone. No editor. RAG does the rest.
Live: https://t.co/bcZ4GfPoXA
Code + write-up in the thread ๐งต
@Aman18007 Each question only sends the closest chunks, at most 10 blocks of about 40 lines, plus the question. The rest of the repo stays out of the prompt. I haven't run the same question on claude to compare token counts.
@Aman18007 Claude reads the files you give it, or searches the repo and opens what it finds. Repo Recall embeds the files you pick, pulls the closest chunks, and the model only answers from those. It points at the file and lines, and you don't have to clone the repo first.