Hi HN!<p>We wanted to introduce our codebase chat that uses a wiki of your repo to do retrieval instead of traditional vector or keyword RAG.<p>We have been pleasantly surprised by the quality of responses. We think creating a language knowledge base for LLMs to do RAG is the future. The way this works is we first write a Wikipedia style article on your codebase complete with diagrams and citations to your code.<p>We’ve run it on some of the most popular repos. If you want us to run in an open source repo please comment below. You can also run it on private repos and the permissions will follow those of the repository on GitHub.
The wikis auto update via a GitHub bot and even show you previews of the new wikis in a diff view.<p>Please check it out and let us know your thoughts.<p>Thank you!
Omar<p>PS Here are some examples:<p><a href="https://wiki.mutable.ai/hashicorp/terraform">https://wiki.mutable.ai/hashicorp/terraform</a><p><a href="https://wiki.mutable.ai/ggerganov/llama.cpp">https://wiki.mutable.ai/ggerganov/llama.cpp</a><p><a href="https://wiki.mutable.ai/ethereum/go-ethereum">https://wiki.mutable.ai/ethereum/go-ethereum</a><p><a href="https://wiki.mutable.ai/NVIDIA/TensorRT">https://wiki.mutable.ai/NVIDIA/TensorRT</a><p><a href="https://wiki.mutable.ai/langchain-ai/langchain">https://wiki.mutable.ai/langchain-ai/langchain</a><p><a href="https://wiki.mutable.ai/ollama/ollama">https://wiki.mutable.ai/ollama/ollama</a><p><a href="https://wiki.mutable.ai/tensorflow/models">https://wiki.mutable.ai/tensorflow/models</a><p><a href="https://wiki.mutable.ai/grafana/grafana">https://wiki.mutable.ai/grafana/grafana</a><p><a href="https://wiki.mutable.ai/OpenAutoCoder/Agentless">https://wiki.mutable.ai/OpenAutoCoder/Agentless</a><p><a href="https://wiki.mutable.ai/unslothai/unsloth">https://wiki.mutable.ai/unslothai/unsloth</a><p><a href="https://wiki.mutable.ai/Dao-AILab/flash-attention">https://wiki.mutable.ai/Dao-AILab/flash-attention</a><p><a href="https://wiki.mutable.ai/vercel/next.js">https://wiki.mutable.ai/vercel/next.js</a><p><a href="https://wiki.mutable.ai/microsoft/vscode">https://wiki.mutable.ai/microsoft/vscode</a><p><a href="https://wiki.mutable.ai/wasm3/wasm3">https://wiki.mutable.ai/wasm3/wasm3</a><p><a href="https://wiki.mutable.ai/deepfakes/faceswap">https://wiki.mutable.ai/deepfakes/faceswap</a><p><a href="https://wiki.mutable.ai/huggingface/transformers">https://wiki.mutable.ai/huggingface/transformers</a><p><a href="https://wiki.mutable.ai/vllm-project/vllm">https://wiki.mutable.ai/vllm-project/vllm</a>
Looks neat. Can you generate on some of my repos (since when I tried it says to contact):<p><a href="https://github.com/theogravity/loglayer">https://github.com/theogravity/loglayer</a><p><a href="https://github.com/theogravity/electron-dl-manager">https://github.com/theogravity/electron-dl-manager</a><p>(Really interested in how it works with this one since it's a monorepo)<p><a href="https://github.com/theogravity/datadog-transports">https://github.com/theogravity/datadog-transports</a>
This is pretty cool, congrats on the launch. Could you run it on Blender? <a href="https://github.com/blender/blender">https://github.com/blender/blender</a>
Please don't request email address when authorizing using github auth. This invariably leads to spam, either intentionally, or accidentally when services don't guard their database / email sending. Instead make this optional and if not entered prompt me to provide an email address once signed up (and do the right email confirmation dance as needed).
Looking at the diagrams of one of your provided examples, <a href="https://wiki.mutable.ai/deepfakes/faceswap">https://wiki.mutable.ai/deepfakes/faceswap</a> I think there is room for improvement<p>I don’t feel that any of those diagrams would really help me if my goal is to dive into that codebase
This is really impressive! Would you mind running it on:<p><a href="https://github.com/Protospace/spaceport">https://github.com/Protospace/spaceport</a><p>This is a member portal for my local makerspace. Having a knowledge base will help other makerspace members contribute to the code.
Any chance you could run this on the Effect codebase? Definitely an example of a project that can be hard to navigate due to its scope and gaps in documentation<p><a href="https://github.com/Effect-TS/effect">https://github.com/Effect-TS/effect</a>
I like the chat interface. It's more dedicated than the universal archival list of past conversations on Anthropic & ChatGPT. Repo is a good unit of work in a developer's mental model. I tried a few chats and answers have been concise.