Skip to main content
Global context gives the Omni Agent background information about your business, so users don’t have to repeat it in each conversation. In AI Hub > Global context, Organization Admins can add global context in Markdown files that the Omni Agent uses in every user’s conversations. Use global context for information that doesn’t fit in a single model, such as what your company does, the terms your teams use, or how users prefer answers.

​
How global context works

You can give the Omni Agent context in three places:
  • Global context — Markdown files that Organization Admins manage from AI Hub. The Omni Agent uses this context in every user’s conversations, for all connections and models the user can access.
  • Personal context — A user’s own context, managed from account settings. See Saving information to memory.
  • Model context — The ai_context parameter in a model file. The Omni Agent uses this context only for that model.

​
Global versus model context

Global context and model context differ in scope: Use global context for information that is true across your business. Use model context for information about the data in one model. For example:

​
When the Omni Agent reads a context file

When you create or edit a context file, you’ll specify when the Omni Agent should read it:
  • Every message — The Omni Agent reads the file’s full content with every message, in every user’s conversations.
  • When needed — The Omni Agent sees only the file’s name and size, not its content. It reads the full file when a question matches what the file’s name suggests it covers. Users can also ask the agent to list the files in context or to read a specific file by name. Use this option for a large reference file, such as a glossary or a runbook, that doesn’t need to be sent with every message.

​
Context and agent behavior

Omni assembles context from various places — global, model, and user levels — and passes it to the Omni Agent. The LLM that powers the Omni Agent makes its own decisions about which context to follow, how to weight it, and how to resolve apparent contradictions. This means:
  • Context is guidance, not strict instruction. The LLM may partially follow, fully follow, or in some cases not follow a given instruction, especially when instructions are complex or contradictory.
  • Global context is not guaranteed to override model or user context. All context is passed together; the LLM decides what to prioritize.
  • Behavior can be non-deterministic. Repeating a prompt or running the same query twice may result in different context being applied, particularly when context is large or contains conflicting signals.

​
Requirements

To create and manage global context files, you need:

​
Creating global context files

  1. AI Hub in the sidebar, then open Global context.
  2. Click New file.
  3. Enter a name for the file, up to 255 characters.
    Use a name that describes the content, such as Product glossary or Regional sales territories. This can help the Omni Agent determine when the file is relevant to a prompt when you opt to have the agent read the context as needed.
  4. Choose when the Omni Agent reads the context: Every message or When needed.
  5. Use the Markdown editor to write the file’s content.
  6. Click Save changes.

​
Editing global context files

  1. Click AI Hub in the sidebar, then open Global context.
  2. Click the file you want to edit.
  3. Update the file’s name, content, or when the agent reads it.
  4. Click Save changes.

​
Deleting global context files

  1. Click AI Hub in the sidebar, then open Global context.
  2. Click the icon next to the file you want to delete.
  3. Select Delete.
  4. When prompted, confirm the deletion.

​
Next steps