Ask Questions with Agent
Get answers from your documents using natural language.
Ask Questions with Agent
The Agent is an AI chat interface that answers questions using your documents. Ask in natural language and get answers with citations.
Steps
- Navigate to Agent in the sidebar
- Type your question in the input field
- Press Enter or click Send
- Watch tool execution (optional — expand tool pills to see details)
- Read the response and click citations to verify sources
What happens when you ask
When you submit a question:
- Query analysis — The Agent decides which tools to use
- Tool execution — Retrieval and analysis tools run (shown as pills)
- Context assembly — Relevant content is gathered from your documents
- Answer generation — The LLM generates a response using document context
- Citation linking — Claims are connected to source chunks
Tool execution indicators
As the Agent works, tool pills appear above the response:
| Indicator | Meaning |
|---|---|
| Amber "executing" | Tool is currently running |
| Green "completed" | Tool finished successfully |
| Red "failed" | Tool encountered an error |
Click any pill to expand and see:
- Tool arguments (what was searched for)
- Tool output (what was found)
- Metadata (timing, chunk counts, etc.)
Reading citations
Citation markers appear in responses (e.g., [1], [cite:abc123:c4]). Click a citation to:
- See the exact source chunk
- Open the source document
- Navigate to the relevant page
Every factual claim should be traceable to a document.
Effective questions
Specific questions work best:
| Instead of | Try |
|---|---|
| "Tell me about the budget" | "What was the Q3 2024 budget for the marketing department?" |
| "What do my documents say?" | "What are the key findings from the Project Alpha report?" |
| "Show me everything" | "List the main recommendations from the 2024 strategy documents" |
Reference entities you know:
- "What is John Smith's role at Acme Corp?"
- "What documents mention Project Paperclip?"
Ask follow-ups: The Agent maintains conversation context. Ask follow-up questions to dig deeper.
Directing tool use
You can guide which tools the Agent uses:
| Phrase | Effect |
|---|---|
| "Use the graph to answer..." | Prioritizes graph tools |
| "Search for documents..." | Prioritizes retrieve_chunks |
| "List all documents about..." | Uses list_documents |
| "Show entity connections..." | Uses knowledge_graph_query |
The Agent chooses tools automatically, but being explicit can improve results.
Search modes
The Agent's retrieval tool supports three search modes:
| Mode | Best for | Example query |
|---|---|---|
| Semantic (default) | Conceptual queries | "What are the key findings about oversight?" |
| Lexical | Exact names, quotes | "Dr. J. Allen Hynek", "Case #1234" |
| Hybrid | Mixed queries | "MK-Ultra mind control experiments" |
The Agent typically uses hybrid mode by default.
Scope and access
The Agent accesses documents based on your current scope:
- Personal scope — Your personal documents only
- Team scope — Team documents (and any shared documents)
Switch scopes in the sidebar to change what documents are queried.