Graph visualizer assistant

The graph visualizer assistant answers questions about the graph you have open in the dashboard. It reads entities, facts, and episodes from that graph and shows each retrieval as it runs.

Use the assistant to:

• Inspect what Zep extracted
• See how two entities connect
• Find which episodes support a fact

You stay in the visualizer.

The assistant is in limited rollout. When it is not enabled for your account, the visualizer does not show Ask graph.

The assistant is read-only. It cannot add, update, or delete graph data. Each conversation targets one graph. Closing the visualizer discards the conversation.

Open the assistant

The canvas occupies the remaining width when the panel is open. The panel width is 20% of the visualizer, with a floor of 320 pixels and a ceiling of 480 pixels.

Ask a question

Type a question. Click Send, or press Enter. Press Shift+Enter to insert a new line.

Before the first question, the panel offers four starter questions:

• Summarize the main entities in this graph.
• Explain how two named entities are connected.
• Show what changed most recently.
• Show the source episodes behind a stated fact.

Selecting a starter puts the text in the input. It does not send the question. Edit the text before you send.

While a turn runs, the input is disabled. Click Stop to cancel the turn. The panel keeps the retrieval steps that already ran and marks the turn canceled.

You can ask follow-up questions that refer to earlier answers in the same visualizer session.

How answers are produced

The assistant grounds each answer in retrieval from the open graph. It does not answer graph questions from model knowledge alone. When a retrieval returns no matches, the assistant states that it found no matching evidence.

Each turn shows the retrieval steps as they run. Steps stay visible after the answer arrives. Each step shows a product name such as Search or Get node, a one-line argument summary, and a one-line result summary. Expand a step to read the arguments and the result.

Selecting a step highlights its entities and facts on the canvas. Selecting an entity name under the answer focuses that entity. An identifier that is not in the loaded canvas is reported as not on canvas. The canvas can be a truncated view of a large graph.

The assistant can use these read tools on the open graph:

• Search
• User summary
• Get node
• Get edge
• Neighbors
• Subgraph
• Episodes

The assistant cannot call write APIs and cannot reach a different graph, project, or account.

Limits and error states

Each question is bounded so a single turn cannot run without a limit:

When retained turns or serialized history exceed a bound, the panel drops the oldest turns until both bounds hold. The panel states that earlier turns left context.

The panel states these outcomes plainly:

• Turn canceled.
• The agent did not reach an answer. Ask a narrower question.
• The tool limit was reached.
• The turn timed out. Ask a narrower question.
• The assistant is unavailable. Try again.
• Rate limit reached. Try again in a minute.
• This graph is not accessible.

The assistant is available only in the dashboard. It is not available through the public API or SDKs. For programmatic reads, use graph search and reading data from the graph.