Master Massive Open-Source Projects Without Burning Your Token Budget
Jumping into a massive, unfamiliar codebase is the ultimate reality check. You try to use an LLM, but you hit file limits almost immediately. You try an autonomous AI agent, and it burns through your API budget before you write a single line of code.
I got tired of fighting narrow context windows, so I built ASTra a repository-scale comprehension tool designed to help you instantly navigate massive Java and Spring Boot projects.
Just point ASTra at any GitHub repository or local directory, and its pipeline structurally parses the entire architecture in minutes. To prove it, I tested it on the JavaParser repository, successfully indexing over 1,750+ classes, nearly 20,000 methods, and 27,000 semantic chunks in just 10 minutes.
From there, you simply ask questions in plain English or drop into one of the purpose-built query modes. Instead of generic LLM guesses, ASTra delivers exact class names, precise line numbers, and real implementation snippets. Whether you use the architecture mode to map out an unfamiliar system, the workflow mode to trace a process end-to-end, or the calls and dependencies modes to hunt down exact usages, it eliminates the endless search for context by giving you immediate, perfectly grounded answers
The project is completely open and actively evolving, and I want developers to push it to its absolute limits. If you want a smarter, significantly cheaper way to conquer large codebases, check out the repository below. I would love to hear your feedback and see what you build with it.
repo - https://github.com/shravan606756/ASTra
demo - https://drive.google.com/file/d/1U3kZZ3bysGIWFf50rIXnQ-UuyociCBes/view?usp=sharing
Jumping into a massive, unfamiliar codebase is the ultimate reality check. You try to use an LLM, but you hit file limits almost immediately. You try an autonomous AI agent, and it burns through your API budget before you write a single line of code.
I got tired of fighting narrow context windows, so I built ASTra a repository-scale comprehension tool designed to help you instantly navigate massive Java and Spring Boot projects.
Just point ASTra at any GitHub repository or local directory, and its pipeline structurally parses the entire architecture in minutes[cite: 1]. To prove it, I battle-tested it on the JavaParser repository, successfully indexing over 1,700 classes, nearly 20,000 methods, and 27,000 semantic chunks in just 10 minutes.
From there, you simply ask questions in plain English or drop into one of the purpose-built query modes. Instead of generic LLM guesses, ASTra delivers exact class names, precise line numbers, and real implementation snippets. Whether you use the architecture mode to map out an unfamiliar system, the workflow mode to trace a process end-to-end, or the calls and dependencies modes to hunt down exact usages, it eliminates the endless search for context by giving you immediate, perfectly grounded answers.
The project is completely open and actively evolving, and I want developers to push it to its absolute limits. If you want a smarter, significantly cheaper way to conquer large codebases, check out the repository below. I would love to hear your feedback and see what you build with it.
repo : https://github.com/shravan606756/ASTra
demo : https://drive.google.com/file/d/1U3kZZ3bysGIWFf50rIXnQ-UuyociCBes/view?usp=sharing
Jumping into a massive, unfamiliar codebase is the ultimate reality check. You try to use an LLM, but you hit file limits almost immediately. You try an autonomous AI agent, and it burns through your API budget before you write a single line of code.
I got tired of fighting narrow context windows, so I built ASTra a repository-scale comprehension tool designed to help you instantly navigate massive Java and Spring Boot projects.
Just point ASTra at any GitHub repository or local directory, and its pipeline structurally parses the entire architecture in minutes[cite: 1]. To prove it, I battle-tested it on the JavaParser repository, successfully indexing over 1,700 classes, nearly 20,000 methods, and 27,000 semantic chunks in just 10 minutes.
From there, you simply ask questions in plain English or drop into one of the purpose-built query modes. Instead of generic LLM guesses, ASTra delivers exact class names, precise line numbers, and real implementation snippets. Whether you use the architecture mode to map out an unfamiliar system, the workflow mode to trace a process end-to-end, or the calls and dependencies modes to hunt down exact usages, it eliminates the endless search for context by giving you immediate, perfectly grounded answers.
The project is completely open and actively evolving, and I want developers to push it to its absolute limits. If you want a smarter, significantly cheaper way to conquer large codebases, check out the repository below. I would love to hear your feedback and see what you build with it.
https://github.com/shravan606756/ASTra
https://drive.google.com/file/d/1U3kZZ3bysGIWFf50rIXnQ-UuyociCBes/view?usp=sharing
dm for repo and demo links, to start contributing, this dumb as server doesnt even allow to attach github links
lol
