RAG for artifacts
Retrieval-Augmented Generation (RAG) for artifacts gives Mark (Optimizely Agent Platform) access to the content in your canvases. A canvas is an interactive document you view and edit within Mark. See Canvas overview for details.
When you ask questions in Mark chat, Mark searches your artifacts and uses relevant content as context. Use RAG for artifacts to ground responses in your previous work, uploaded files, or agent outputs. Mark draws from your canvas content rather than general knowledge.
Mark respects artifact permissions during retrieval. Mark includes artifacts shared with you alongside your own.
For an overview of how RAG works across Mark, see Retrieval-Augmented Generation (RAG) overview.
How RAG for artifacts helps you
RAG for artifacts lets you and your team do the following:
- Find information across past work – Search agent outputs, uploaded files, and stored content without leaving chat.
- Build on previous results – Reference outputs from earlier workflow runs when starting a follow-up task.
- Ground responses in your content – Mark draws from your actual artifacts rather than general knowledge alone.
Verify RAG for artifacts is active
Optimizely automatically provisions RAG for artifacts for eligible Mark instances. Administrators do not need to take action.
Confirm RAG for artifacts is active for your instance:
- Go to Settings > RAG Configuration.
- Verify that your Mark Artifacts instance displays Active on the Configurations tab.
The Backpop Jobs tab displays the status of indexing jobs that make your existing artifacts available for retrieval.
Use RAG for artifacts
Mark searches your canvases automatically when you ask a question in chat. Ask questions without any special syntax.
In Mark chat
Ask your question in Mark chat. Mark searches your canvases and includes relevant artifacts as context. When Mark references an artifact, the response includes the artifact name and the space it came from. This lets you locate the source directly on the Artifacts page overview.
In a specialized agent
Use the {retrieval:} and {retrieval+:} directives in your prompt template to retrieve artifact content during a specialized agent run. See Prompts for specialized agents.
RAG limitations
The following limitations apply to RAG for artifacts:
- Optimizely supports one RAG for artifacts instance per Mark instance.
- Mark retrieves artifact titles, summaries, and similarity scores. To read the full content of a retrieved artifact, open it from the Artifacts page.
- Retrieved artifacts reflect the state of your canvases at the time of the last index. Content added after the most recent index is not available for retrieval until the next index runs.
- Optimizely does not guarantee retrieved artifact content is current or accurate. Treat it as background context and verify against the source artifact before acting on it.
Supported artifact types
RAG for artifacts indexes the following content from your canvases:
- Mark chat outputs saved to a canvas.
- Files you uploaded to canvas.
- Output from workflow and specialized agent executions.
- Stored memory artifacts.
- Artifacts that other users shared with you.
Note
If you use Opti ID, administrators can turn off generative AI in the Opti ID Admin Center. See Turn generative AI off across Optimizely applications.