We built Forums because we got tired of the same loop: find an open source library, docs are outdated, end up digging through source code for hours.
Or chatting with claude and friends... just to see that they too have stale data due to their knowledge cutoff.
So we thought: what if you could just ask the repo directly?
Forums lets you ask any public GitHub repo a question and get source-backed answers. We clone the repo in a Vercel Sandbox, and the agent greps through the code, and finds real answers. No hallucinations. No stale docs. Just the source.
What you get:
Ask - Ask questions, AI explores the code, shows you exactly where things live
MCP server - Plug into Claude Code, Cursor, or any MCP client. Ask about external repos without leaving your editor
CLI + skill - Run commands against any public repo without cloning locally. And without signing in to Forums at all... let it rip.
We use it constantly. Integrating a new library? Debugging weird behavior? Faster than asking an LLM that was trained on a year-old version of the code.
I came across Forums — the AI‑powered Q&A tool for GitHub repos — and love the concept of turning code into meaningful answers.
One thing I noticed is that the product could benefit from stronger visual clarity and brand identity — especially in the homepage hero, UI hierarchy, and brand polish. This matters because users decide whether to engage in the first few seconds based on how professional and clear the interface feels
Last week, my manager asked me to fix a legacy service, and it took me three whole days just to figure out where to start. If I had this tool back then, I probably could've done it in three hours. For a newcomer like me, this is basically a cheat code.
Brilliant idea! Does this use the latest commit or cached data? (Some apps I've tried have stale code issues. Also, how does it perform on large-scale repos? Can the AI stay accurate when navigating thousands of files?
Wait, would it be possible to become an MCP? I mean, I can see many possibilities here haha
This is super useful — being able to ask questions with actual source-backed answers solves a real pain. Curious how you handle large repos and context limits?
It sounds super handy, but I’m mainly wondering about the privacy side like, if I use it on a private repo, is the code kept secure and deleted after the session? Also, does it need a lot of permissions to access my GitHub?
The fact that it provides 'source-backed' answers is great because AI can sometimes hallucinate code. I wonder if it actually links to the specific lines of code it used for the answer so I can double-check everything myself.
This sounds like a total life-saver for when I have to dive into a massive, messy codebase for the first time! I’m really curious if the AI is smart enough to understand the logic across different files, or if it just searches for specific keywords.
Well done on the launch. Awesome idea. Identifies a gap in servicing a user's needs / pain point and solves it. I'm developing something on the same principles ATM that may get a formal release announcement some day.
Congrats on tha launch!, actually i also had an idea to build a side project like this but thought ai tools like cursor and claude code can already do this, how is this different, would love to know about that
I love this. But I'm concerned about requesting access to private repos. They are set as private for a reason. Mind explaining why that choice? In any case, I'm supporting the idea and therefore the launch. I might not use it because of the above, but it is certainly something needed and useful.
hi. how is this different from GitHub discussions that's already there as a native feature?
Real-time exploration is much more efficient than traditional searching, and the code structure is clear at a glance. When parameters with the same name but different meanings appear in different files, how does the system ensure the accuracy of interpretation?
Integrating a new library is much faster than reading the documentation. It's like having a friend who understands code by your side. If there are "//TODO" or deprecated comments in the code, how will the AI remind users to pay attention?
gm PH 👋
We built Forums because we got tired of the same loop: find an open source library, docs are outdated, end up digging through source code for hours.
Or chatting with claude and friends... just to see that they too have stale data due to their knowledge cutoff.
So we thought: what if you could just ask the repo directly?
Forums lets you ask any public GitHub repo a question and get source-backed answers. We clone the repo in a Vercel Sandbox, and the agent greps through the code, and finds real answers. No hallucinations. No stale docs. Just the source.
What you get:
Ask - Ask questions, AI explores the code, shows you exactly where things live
MCP server - Plug into Claude Code, Cursor, or any MCP client. Ask about external repos without leaving your editor
CLI + skill - Run commands against any public repo without cloning locally. And without signing in to Forums at all... let it rip.
We use it constantly. Integrating a new library? Debugging weird behavior? Faster than asking an LLM that was trained on a year-old version of the code.
Would love to hear what repos you try it on.