Ovellum for AI agents#
Documentation now has two audiences: the people who read it, and the agents that read and increasingly write it. Ovellum treats the second audience as a first-class concern, not a plugin. Three things come for free.
Read-ready#
Every build emits machine-readable companions next to the HTML, following the llmstxt.org convention:
/llms.txt— a link-first index of every page./llms-full.txt— the whole corpus as one Markdown stream.- a
.mdmirror of every page, byte-for-byte the Markdown behind the HTML.
So an agent reads your docs as clean Markdown instead of scraping rendered
pages. On by default; see site.ai.
Agent-drivable#
An agent can drive Ovellum without parsing human output. build, check, and
diff speak --json with stable exit codes, and
ovellum mcp runs a Model Context Protocol
server exposing the operations as tools. Setup is a copy-paste away —
a scaffolded AGENTS.md and a Claude Skill tell an
agent how to use it.
Safely editable#
This is the part no other docs tool offers. Because Ovellum's
hybrid merge engine keeps hand-written prose inside
protected zones, an agent can contribute
real prose to generated docs and have it survive the next regeneration —
exactly the guarantee a human editing between @manual markers gets. Over MCP,
that's the ovellum_write_zone tool.
And when a refactor moves the ground under that prose, it isn't lost:
ovellum diff flags the likely rename, and
anything that can't be re-placed is quarantined to
.ovellum/orphans/ for review rather than silently
dropped.
Read-ready, drivable, and safely editable — the same anti-drift contract that keeps human prose and generated docs honest, extended to agents. Start with the Automation & AI agents guide.