AI Assistant Integration

[fengtality]

I'm seeing more and more documentation sites add an Ask AI widget. Given the current state of LLMs and new developments like MCP, I think mkdocs-material should consider adding this widget feature or some other UI for an AI chat interface.

In my opinion, the documentation provider can hook up that interface with their agent of choice, so mkdocs-material should try to provide a simple, generic interface that can handle streaming conversation.

[squidfunk]

Thanks for noting! We already wrote about this in our latest blog post. We'll consider adding support for AI-related features once our current foundational work reaches stability ☺️ Note that we're currently focusing on engineering topics, but we're close to a release.

[fengtality]

that post was published a year ago 😭 Michael Feng ***@***.***

…

On Tue, Sep 2, 2025, 11:05 AM Martin Donath ***@***.***> wrote: Thanks for noting! We already wrote about this in our latest blog post <https://squidfunk.github.io/mkdocs-material/blog/2024/08/19/how-were-transforming-material-for-mkdocs/#large-language-models-llms>. We'll consider adding support for AI-related features once our current foundational work reaches stability ☺️ Note that we're currently focusing on engineering topics, but we're close to a release. — Reply to this email directly, view it on GitHub <#8433 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AANWHVTEYUP26D6G66T3OG33QXL5JAVCNFSM6AAAAACFOBRK6OVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTIMRYHA2TCMI> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>

[squidfunk]

We're close to our initial release. Once that's out, I think it will be very clear why this took us more than a year, as this has been a giant undertaking. Also please keep in mind that we're not a VC funded company with 20 employees – we're a bootstrapped team of 3 people with 1 developer. Additionally, we decided to wait until the AI community settles for standards. MCP looks more promising than all RAG-based AI chatbots I've seen, which IMHO provide mediocre results. Dislaimer: I haven't closely followed the AI trends in the past months, as we've focused on deep engineering topics, and not another thin AI-wrapper technology.

Regardless, we're very excited for the upcoming weeks!

[fengtality]

thanks for the update - excited to test it out! fwiw, we are similarly bootstrapped with only 2 devs and have found that AI has supercharged our team's productivity and let us take on more/riskier projects. Michael Feng

…

On Wed, Sep 3, 2025, 12:10 AM Martin Donath ***@***.***> wrote: We're close to our initial release. Once that's out, I think it will be very clear why this took us more than a year, as this has been a giant undertaking. Also please keep in mind that we're not a VC funded company with 20 employees – we're a bootstrapped team of 3 people with 1 developer. Additionally, we decided to wait until the AI community settles for standards. MCP looks more promising than all RAG-based AI chatbots I've seen, which IMHO provide mediocre results. Dislaimer: I haven't closely followed the AI trends in the past months, as we've focused on deep engineering topics, and not another thin AI-wrapper technology. Regardless, we're very excited for the upcoming weeks! — Reply to this email directly, view it on GitHub <#8433 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AANWHVWM6Y6KIQPBNKVE5ML3Q2H5TAVCNFSM6AAAAACFOBRK6OVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTIMRZGI4TQNQ> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>

[squidfunk]

Don't get me wrong, LLMs are super helpful hammering out repetitive things like tests, docs, or simple prototypes. I was primarily referring to hosting chat bots for users to interact with your documentation.

[fengtality]

My recommendation would be to start with a thin wrapper where users can add Anthropic/OpenAI API key and a model that responds to questions in a chat interface embedded on the site. I think the work on the mkdoc-material side is on the chat interface side of things, whereas how model is designed (whether it uses RAG or some framework) should be left up to the user.

Depending on how you want to handle tool use and streaming conversations, the chat interface could get complex but I would start with a simple chat supporting StreamableHTTP.

[PatrickFreyETAS]

I have been investigating AI support lately for some product where we have the docs online.
The most promising solution we have found was Kapa.ai. They specialize on AI assistants for technical documentation.
https://www.kapa.ai/

They are a commercial vendor but very approachable.
Could be a win-win for both MkDocs-Material and Kapa.ai to have their AI assistant integrated with MkDocs.
I do not know if Kapa.ai has some kind of free plan for projects such as MkDocs-Material.

[squidfunk]

Since none of those docs AI startups actually make any money, my guess is that this AI bubble will pop very soon. Any solution or integration of LLMs that we come up with cannot possibly depend on some startup that might be gone in a few months. That's just a waste of engineering effort. If kapa.ai wishes to support MkDocs, they can do so through its plugin architecture. Intergrations are fairly straightforward and do not depend on the internals of Material for MkDocs.

[fengtality]

the AI bubble isn't popping anytime soon but I agree with Martin that there shouldn't be any 3rd party technologies needed to support AI. IMO, all that's needed is letting the docs provider add a model API key and and an interface that lets the model answer user questions.

…

On Fri, Sep 5, 2025, 1:29 AM Martin Donath ***@***.***> wrote: Since none of those docs AI startups actually make any money, my guess is that this AI bubble will pop very soon. Any solution or integration of LLMs that we come up with cannot possibly depend on some startup that might be gone in a few months. That's just a waste of engineering effort. If kapa.ai wishes to support MkDocs, they can do so through its plugin architecture. Intergrations are fairly straightforward and do not depend on the internals of Material for MkDocs. — Reply to this email directly, view it on GitHub <#8433 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AANWHVTQ5XDLCC3QGFKLA7L3RFCXHAVCNFSM6AAAAACFOBRK6OVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTIMZRGU3TENI> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>