LLM Instructions
LLM instructions guide AI agents working with your collections and globals. Payload core manages configured Markdown instructions and additional instructions saved through the Admin Panel. The feature works without the MCP plugin.
Both the CLI and MCP plugin return configured instructions combined with saved additional instructions as Markdown. Only getCollectionSchema and getGlobalSchema include them, so agents should read the schema before creating or updating content. Instructions are omitted when both sources are empty.
Configuration
Set llmInstructions on a collection or global to provide read-only Markdown instructions. Use the root llmInstructions option to control instruction management:
Editing instructions
Open Edit LLM instructions from a collection's list menu or a global's edit menu. The hidden LLM Instructions collection has one configuration-owned entry per visible collection or global. Its Additional instructions tab saves editable instructions; System instructions (read-only) displays the configured Markdown and cannot be edited through the Admin Panel or API.
Editor
Lexical uses a dedicated instructions editor with paragraphs, H2–H4 headings, bold, italic, strikethrough, ordered and unordered lists, and a fixed toolbar. Its feature set is independent of the root editor's features. Set llmInstructions.editor to a separate rich-text adapter to override it.
Access control
Authenticated users can read saved instructions for collections and globals they have read access to. Editing also requires update access to the target and permission through llmInstructions.access, which defaults to users of the admin auth collection. Target access is evaluated at the entity level, as in the Admin Panel.
Disabling management
Set the root llmInstructions option to false to disable the management collection and menu items. Configured collection and global instructions are still included in CLI and MCP schema responses.
Was this page helpful?