---
agentTools:
  siteIndex: https://docs.readme.com/llms.txt
  siteSearch: https://docs.readme.com/agentSearch?query=term1%20term2
  projectIndex: https://docs.readme.com/main/llms.txt
  projectSearch: https://docs.readme.com/main/agentSearch?query=term1%20term2
---

# Custom LLMs.txt

We’ve built support for a [custom LLMs.txt file](https://docs.readme.com/main/docs/ai-discoverability#content-discoverability), so git-connected docs sites can serve their own hand-authored version instead of the automatically generated one. You can switch back to the generated version anytime from your project settings.

### New Features & Improvements

**Docs**

* Required parameters in API Reference now come pre-filled with example values from your API definition, so Try It works without manually filling in every field.

**Enterprise**

* Enterprise admins can now view audit logs from the project page as well as in the Enterprise dashboard.

**MCP**

* MCP tools now expose human-readable titles (like "List API Endpoints" instead of `list-endpoints`) so AI clients can display them more clearly.

<br />

### Bugs Eaten (by Owlbert) 🐛

**Admin**

* Dashboard quick-search keeps your keyboard selection in place while results are still loading.
* SSO users can now complete email verification even when their account was already verified through SSO.
* Connecting a chat notifications workspace that's already connected elsewhere now shows a clear error you can retry from.

**Enterprise**

* Child projects in an enterprise group can now opt out of the group's default LLMs.txt setting.
* Long lines in enterprise audit log code diffs now scroll instead of getting cut off.

**Editor**

* Branch reviews now show a callout as edited, not deleted, when only its title changed.
* Password managers no longer auto-fill the page description field in the editor.
* AI agent edits to a page now show up in real time for others editing the same page.

**AI**

* AI agent skill listings are now readable by agents that follow the current discovery spec.
* Ask AI model picker now shows readable model names instead of raw internal IDs.
* Ask AI widget now shows an error on response failure.
* Ask AI defaults to medium length when no specific length is configured.
* Ask AI's suggested questions now come back in the language you’re browsing in instead of defaulting to English.

**CLI**

* GitHub Action now validates and uploads OpenAPI/Swagger definitions provided as a remote URL.
* Docs linter now catches duplicate-slug errors on parent pages that use a folder-plus-index-file structure instead of a sibling file.