Docsy joins the Linux Foundation and starts scoring docs for AI agents

A documentation site used by Kubernetes, OpenTelemetry and gRPC is quietly being rebuilt for a reader who does not scroll. The New Stack reports that Docsy, the Google-built theme for the Hugo static site generator, is moving to the Linux Foundation and is adding features aimed at AI agents, with a plan to score how easily any site built on it can be read by one.

Erin McKean, senior developer relations engineer at Google and a member of the Docsy steering committee, announced the move in a keynote on Wednesday at the Linux Foundation's Open Source Summit Europe in Prague, speaking afterwards to The New Stack's Paul Sawers. By the end of 2024, some 2,200 projects were using Docsy, with adopters across the Cloud Native Computing Foundation including Kubernetes, OpenTelemetry, gRPC and Jaeger. First announced by Google in 2019, Docsy has grown up inside the open-source side of Google and will now sit closer to the projects already using it.

What Docsy ships for AI readers today

McKean said in her keynote that documentation is read now by AI assistants as well as by people, and that the project has been adding features for that second audience for the last five months. Since Docsy 0.15.0 in May, a site can generate a flat Markdown copy of each page alongside the HTML, plus an llms.txt file at the site root that lists what the site publishes. Both features are opt-in and experimental.

Version 0.16.0 in July added an upgrade guide written so an AI assistant can follow it, with the conditions, steps and checks spelled out. Version 0.17.0 in August went further: a site that enables llms.txt now also adds a hidden directive at the top of every page that points visiting agents to the llms.txt index. McKean's framing to Sawers: "You can redirect your LLMs and agents to the text that tells them how to use the project."

The next item: a score for how readable your docs are

Next on the Docsy roadmap are "AF" scores, meaning agent-friendly documentation scores, designed to give maintainers a measurement rather than a feeling. "You won't have to guess," McKean told The New Stack. "You can measure how agent friendly your docs are." Neither the keynote nor the interview gave the metrics the score is made of, so the shape of the benchmark will arrive with the release.

The second argument McKean made is older than the AI story: good docs reduce the questions landing on maintainers, which matters whether the next reader is a developer or an agent. The practical read for a team maintaining an open-source project on Docsy is that the project's next release gives you two quiet opt-in flags you can turn on in a config file, and a score to improve against once "AF" ships. For a team writing docs for a proprietary stack, the same theme still applies: a Markdown copy and an llms.txt index are the two shortest paths an AI assistant has to the page you wanted it to read.

Source

Primary source: Docsy. Report: What's up, Docsy? Google's docs project joins the Linux Foundation as AI agents become readers on The New Stack, by Paul Sawers.