back
31 comments
We need to make an effort to distinguish “this is a thing for humans” vs “this is a thing for bots” in our naming IMHO. In that respect, “open wiki” is not such a great name. “Agent Wiki” or similar would be better.

Without such a qualifier, “wiki” carries a strong connotation of (usually collaborative) human involvement. That’s literally what it’s famous for.

Thats just my $0.02 on the naming. I definitely think it’s a worthwhile idea. All the best.

I recently noticed people think "wiki" is a set of deeply interlinked documents. But to me it means ability to edit without logging in.
I don't know of any wikis that let you edit without logging in. That is simply not viable, it'd be like trying to live without an immune system.
The original C2 wiki did that for two decades.
Wikipedia?
You try that in 2026 and not only is your changes getting rolled back but your IP address is likely banned in advance anyway. Though to be fair I hear that even logged in your changes get rolled back, something about "reddit style moderators power tripping".

Though on the naming definition I think wiki is a good term for interlinking documents about _something_. I often associate the term wiki with game wikis, which more often than not at least require some form of account to edit. Wikipedia feels like an outlier (both in not being specific to any topic as well as technically allowing anonymous edits) even if it presumably is the first/originator of the name!

Yeah, Wikipedia is famously hostile to new contributors. It's a wonder they still continue existing and also have the gall the beg for donations. Imagine if reddit begged for donations?

I don't even really think about wikis as a set of interlinked documents moreso just as a repository of information. In the context of games whose communities created wikis, at least. Usually the discoverability of and interlinking of documents in said community wikis is pretty piss-poor though.

This is mostly a thin clintypescript wrapper around the prompts.

This could have been a SKILL

it's langchain, what did you expect
What does this do better than just asking your agent to "write docs" or a more robustly defined prompt/skill?
Have a look at the prompts in the GitHub [0]. It defines a System Prompt and specifies the documentation structure. This would allow you to switch coding agents, instead of relying on how your coding agent interprets the command "write docs".

[0]: https://github.com/langchain-ai/openwiki/blob/main/src/agent...

That's just cruft unless there is a benchmark demonstrating it's actually better than just asking an agent to write docs and/or using one of the thousands of document-writing skills like Anthropic's doc-coauthoring.
I swear most of these tools are made for the sake of it…

While good old prompting is often better than plan mode or superpower skills.

I've had a number of people send me tools like this at work, and easily 50% of them can't answer basic questions like "what's the reason someone would adopt this tool" or "how do you know that it will achieve its stated goals". Agents are good at reading code, I can't imagine what the point of autogenerating agent context could be if it's not showing demonstrable cost benefits.
> I can't imagine what the point of autogenerating agent context could be

I have a fork of the OpenAI Codex repo at https://github.com/gitsense/smart-codex that shows why you may want to autogenerate agent context.

I makes navigating a 4000+ file repo extremely context efficient. It is important to note that the goal is too keep the context as clean as possible and not necessary speed.

This is what we do. The same agent writing the code can also write the docs.
maintaining an LLM wiki has been a lot more effort than I thought, at least if we are trying to maintain a high quality in structure and writing comprehension (for easier lookups both for the agent and human). Are people just shotgunning their agent wikis or how
> maintaining an LLM wiki has been a lot more effort than I thought

Same here. Wikis start out good, but either devolve in a journal-y mess after a while and many updates, or require constant expensive rewrites. (I didn't use the software of the OP.)

Wikis start out good, but get stale too quick and become useless or worse confusing.

I’ve experienced this over and over again to strongly believe it.

I genuinely wonder if throwing LLMs at this problem would solve it at least to some extent. Make a LLM agent whose sole purpose is to act as a librarian. It periodically reviews _all_ of the wiki and validates them against codebases, newer docs, anything. Whatever it finds, it should be allowed to intelligently quiz the team/dev whether something is right/stale/wrong and updated it accordingly.

If one tolerates that toil - answering questions of a library bot, would it result in a usable wiki base?

I think writing the instructions for such an agent would be almost as much work as maintaining the wiki and learning your preferences yourself (and both may be coupled intrinsically unless you already manage a large enough notes system). The crucial problem is that the agents lack 1. taste and 2. the ability to know what is good for them. I've bounced off trying these like three times over the last couple of years, twice after Feb 2026, simply because it requires way too much toil that would be better (IMO) put into maintaining your own knowledge base, where at least that toil would result in learning
In short, yes. The effort will amortize for a larger team over longer time period.

https://openai.com/index/harness-engineering/

Unless it's about motivation and other things that can't be inferred from the code (and comments on such are missing), just ask the agent. Give it an LSP or code intel MCP to do it better.
I think it's own sub-dir OpenWiki serves as an example of the sort of output you might get:

https://github.com/langchain-ai/openwiki/blob/main/openwiki/...

I agree with others this seems somewhat over-engineered; you can get similar results with a good prompt/skill; I guess the rest of the implementation here is intended as an agent-maintainer.

Needs actual evidence that this is better than not having documentation, only AGENTS.md and letting agents explore.

Having verbose, duplicated, incorrect, outdated or incomplete docs often leads to worse performance for AI

Is this the tool that blasted away all the wiki pages for Azure in the repos?
How does the ability to search code snippets and symbols compare to Codegraph?
extension for pi.dev https://github.com/barvhaim/pi-openwiki