hey I'm from the SWE-grep team - feel free to ask me any questions :)

No question just wanted to say good job and thanks as a user. Same with deepwiki and codemaps.

Did you intend to answer them, or you just wanted the questions?

Are you actually using grep here? How much data are you searching?

no - grep is just the closest analogy/use case that we have for it. if we end up releasing the CLI it should be as handy and nobrainer as using ripgrep

idk what you expect from a question about "how much data". its tool based search. its a lot.

I'm just learning about agentic search so I'm a bit adrift.

One of my side projects is a full text index for pattern search, and I'm trying to understand how it might fit with that. You mention tool call overhead, but is that a significant part of the latency in the multi-turn scenario, or is it the coding agent being forced into a serial processing pattern?

hey sorryjust saw this. i do think its majority serial processing, BUT, parallel calling the same tools also gets issues that i honestly havent spent the time to dig into (something something locks and threading). all i know is ive been stuck in very very super slow/tool calls myself in Windsurf/other AI IDEs and that was a drag.

for another take on latency attribution see https://x.com/silasalberti/status/1979310181424206143

this would be useful outside of coding. could you release a benchmark so we can have more models tuned for this?

What base model did you use?

Hey I'm Silas, I worked on this. Lmk if you have any questions!

Hey Silas, currently deepwiki is not working for me. Yesterday it was working great and I found it really helpful, today I only see the loading page with skeletons with every prompt.It's slow and most of the times the results never come in. Previously i could see it's chain of thought now there's just the loading page with skeletons. Just wanted to let you know.

This seems like a really cool project and could be really useful in getting to know a new codebase. How do you ensure that the output in the wiki does not contain hallucinations?

When working with LLMs you can never exclude hallucinations entirely but we‘ve been carefully tuning the system over time and found it to be pretty high signal! The reason we display the code snippets is to make it easy to double check with the source

Can you "chunkify" output so that you can rate different elements independently? Like "This part of the answer is totally cool" and "Wait. This part right here includes a hallucination."

Then allow for feedback to be provided if an issue is spotted.

This might be a bit off-topic but is this custom-built documentation or did you use a template? I've been on the hunt for pretty documentation tools but have only come up with Mintlify.

This is entirely custom built but we‘re also fans of Mintlify. We use them for docs.devin.ai

Is it custom all the way down or based on something like starlight? I really like Mintlify but $200 per doc site hurts for internal tools.

I can see a lot of merit in custom building though especially with making it easier to dump it into `llms.txt` or exposing a search for AI. Hoping that's where DeepWiki is headed :)

Hi Silas!

Are there any plans to allow running DeepWiki against private non-GitHub repos (e.g. GitLab, Azure DevOps, etc) and keep the output private?

Great work Silas! Can this also be trained on externalized sites, such as human-written docs, our web site, public-facing Google docs, etc?

Currently it uses only the codebase itself. However we‘ve been thinking about adding other source like docs. Can’t promise any timelines yet

Another idea: allow more "chunkified" approvals, and verbatim feedback:

"This part is correct, but this next paragraph? You're hallucinating / misinterpreting. What you really need to say is 'x'."

Also, I suggested that feedback also get scored in some way. For instance, what if someone is sending malicious feedback?

How do you discourage or de-weight someone downvoting good answers while upvoting wrong answers?