Mutable.ai is a living AI expert on your codebase, that is: 1) aware of your entire codebase, 2) can answer almost any question at a human level of quality.
Try it out on an example here https://wiki.mutable.ai/langchain-ai/langchain
Mutable.ai v3 is here. Mutable.ai is aware of your entire codebase and can answer questions with an uncanny human-like quality.
Generic LLMs have two flaws:
1) They currently have context windows that are too small to accommodate most codebases, let alone your entire org's codebases.
2) They need to reason immediately to answer any questions without thinking through the answer "step-by-step".
Mutable.ai does it's homework by writing an entire Wikipedia style article on your codebase, this wiki serves as an intermediate reasoning step that allows it to answer questions more intelligently.
We also leapfrog traditional RAG by giving an LLM a "language map" of your entire codebase, it turns out language models love being given language and not a bunch of vectors or keywords!
We are introducing evals demonstrating how much better our chat is with this approach, more on this soon!
Congrats on the launch! Mutable.ai sounds impressive. A quick question - does it handle multiple coding languages? And how does it deal with different coding styles or conventions within a project? Excited to see where this goes!
A really cool project! This looks like a great time saver @omar_shams. Is there any more information about how this works? I'm wondering where the codebase is sent and how the data is stored, especially in context of private repositories.
Hey @dleszczynski our security policy is here https://mutable.ai/security and we're working on SOC. We follow all best practices, your data is stored on AWS S3 buckets, all interactions by our small team are logged, we used 2FA and never look at your data without your express permission
@omar_shams Thank you, looking forward to see where you guys take this product :) I might ask my team if they'd like to try it out, we are cooking a new open source project.
@omar_shams Great idea. I think it would make more sense to show the wiki article first. So that I can decide if it's useful. But I understand that at the MVP stage, you can't make it perfect. Also maybe instead of asking to sign in to GitHub, you can rather make a PR. That would make more sense for me. Since I store documentation as .md files in the source code rather than in the wiki. That way my IDE AI tools sees the documentation and can reason about the code in a better way. # my 2 cents
Mutable.ai