Using Kaimon for documentation #58

New issue

Open

opened 2026-07-09 05:57:35 -07:00 by BambOoxX · 1 comment

BambOoxX commented

2026-07-09 05:57:35 -07:00

(Migrated from github.com)

This is more a question than an issue.

When developing the documentation (in my case with Documenter) of a package I often have documentation generation errors due to lacking references or docstrings.
I figured there is something Kaimon could be of help there to streamline this using an agent.
What good practices would you recommend Kaimon for documentation, in particular, how to properly link the output of the generation and the agent through Kaimon.

This is more a question than an issue. When developing the documentation (in my case with `Documenter`) of a package I often have documentation generation errors due to lacking references or docstrings. I figured there is something Kaimon could be of help there to streamline this using an agent. What good practices would you recommend Kaimon for documentation, in particular, how to properly link the output of the generation and the agent through Kaimon.

kahliburke commented

2026-07-10 22:19:03 -07:00

(Migrated from github.com)

It's also been helpful to make sure that the build process is fully automated including any figure/image/animation generation tasks which run through CI.

I would not say it's been anything especially formal, and it's taken a certain amount of dedicated time and effort focused on this. I also wouldn't claim that the docs are perfect, but I think the results are higher quality than many open source projects.

If you want to discuss some more we could discuss your specific use cases here or on Discourse?

@BambOoxX I've used Kaimon itself (with Anthropic LLM models) to develop the documentation for this project (https://kahliburke.github.io/Kaimon.jl/dev/) as well as for https://kahliburke.github.io/Tachikoma.jl/dev/ and https://kahliburke.github.io/KaimonSlate.jl/dev/. As you can see they share the same DNA. This is using `Documenter` and `vitepress` plugin. My approach has been to use the semantic indexing and search functions, along with some time spend on the overall design and goals, and set the agent at the task having it utilize the indexing services to learn the code and to properly extract the correct sections. It's also been helpful to make sure that the build process is fully automated including any figure/image/animation generation tasks which run through CI. I would not say it's been anything especially formal, and it's taken a certain amount of dedicated time and effort focused on this. I also wouldn't claim that the docs are _perfect_, but I think the results are higher quality than many open source projects. If you want to discuss some more we could discuss your specific use cases here or on Discourse?