What an agent sees determines how well it answers. The context engine is how you control that: by referencing files, notes, and folders with @, you give agents a clear view of your sources and specific instructions with direct references—and that means dramatically better citation quality.
What Are @ Symbol References?
Typing @ in the chat attaches items from your workspace to a prompt. Typing @ opens a picker where you can choose:
- PDFs and other files
- Notes
- Folders
- UDOCs / USRCHs
What Can Be Referenced?
- Files: @report.pdf, @contract.docx
- Folders: @case-files/, @research-materials/
- Notes: @note:id or similar identifiers
- UDOCs: @draft-v2.udoc
You can reference multiple items in one prompt. The agent then grounds its answer in those items instead of general knowledge. That reduces hallucination and makes citations traceable—often verifiable with one click to the exact location in the source.
How References Affect Output Quality
Research shows that when models reason and cite at the same time, accuracy drops sharply. The main problem is invented quotes and wrong sources. When you use @ references:
- Agents read from the actual documents you selected
- Claims link to specific pages and passages
- You can verify each citation in the source file
Prompts that include @ references tend to produce more reliable output.
Example Prompts: Simple to Expert
Simple – one @ reference:
Summarize the key findings in @market-report-2025.pdf
Intermediate – multiple references, multi-step:
Compare @report-a.pdf and @report-b.pdf. Create a synthesis table with exact quotes and page citations where they agree and disagree on ROI timelines.
Expert – full folders, note:id, multi-step:
Search the folder @case-files/ for holdings on implied warranty of merchantability. Create notes with exact quotes from each opinion. Find distinguishing factors between @client-fact-pattern.pdf and @smith-v-jones-2023.pdf. Flag the exact language I need to address and create a new UDOC that uses the quotes you find to explain your findings.
All Ways to Add Context to the Chat Window
- Drag and drop – Drag files or folders from the sidebar into the chat
- @ reference – Type @ and select items
- + button – Click + and choose items from the workspace
- Academic Database – When viewing search results, use actions to add sources and bring them into the chat
What Agents Do with Better Prompts
With solid context, agents can:
- Extract findings across multiple documents
- Build synthesis tables with citations
- Flag contradictions between sources
- Map claims to evidence in your files
Outputs are evidence-based and traceable. You can accept, edit, or reject suggested changes in the UI before applying them.