# AI features

aardvark is built for two audiences at once — the people reading your docs and the agents
acting on them. The reader features add an assistant; the agent features make your site
discoverable, queryable, and verifiable by automated clients. Everything is in one place
here.

## For readers



A built-in "Ask AI" chat panel for your readers — plus a full analytics dashboard (Top Questions, Coverage Gaps, and more) for you.




The managed metering proxy behind the built-in assistant — a prepaid balance with Stripe card-on-file and auto-top-up.




Opt-in, cached OpenRouter features — generate frontmatter, example API responses, and Claude Code skills from your docs.



## For agents



Run `vark serve` to expose a live MCP server and serve each page as Markdown via content negotiation — the same tools work in the browser through WebMCP.




A checklist for making your site agent-ready — what aardvark publishes on every build, and the few steps only you can do.




Always-on discovery endpoints — an MCP Server Card, OAuth/OIDC metadata, `auth.md`, and DNS-AID records — emitted by every build.




Publish a key directory so your site can identify itself when an agent sends signed requests on its behalf.






The files agents read most — `llms.txt`, `llms-full.txt`, and the Agent Skills index — are written by every build; see [Generated files](/llms-and-sitemap/). The `vark ai-enrich` and `vark web-bot-auth-keygen` commands are in the [CLI reference](/cli/).