Skip to main content

Using Forward AI

Opening Forward AI​

Click Forward AI in the left navigation. The page opens in the currently selected network and uses the latest processed snapshot.

Forward AI page with the entry highlighted in the left navigation

Page layout​

The Forward AI page is split into three areas:

AreaWhat it shows
Conversations panel (left)Your past conversations on the current network, sorted most-recent first. Click a conversation to reopen it; right-click for rename and delete.
Transcript (center)The current conversation: your prompts, the agent's replies, and the prompt input at the bottom.
Plan & tools panel (right)The agent's plan for the active prompt and the tool calls it executed, with their results.

NQE results and other multi-row tool outputs render in a carousel below the relevant assistant message so you can page through them without leaving the conversation.

Conversations panel, transcript, and Plan & Tools panel on the Forward AI page

Starting a conversation​

A new conversation displays Get started with these prompts — example prompts grouped under:

  • Troubleshooting — path tracing, reachability checks, blocked-traffic diagnosis.
  • Vulnerabilities — CVE exposure, vulnerable paths, prioritization.
  • Compliance / Audit — config consistency, unused rules, direct internet exposure.
  • Inventory — EOL devices, BGP session audits, host discovery.

Click an example to drop it into the prompt input; replace the <placeholder> values with real ones from your network, then press Send. To skip the suggestions, type your own prompt and press Send.

Empty Forward AI conversation showing the four prompt categories

What happens after you send a prompt​

  1. The conversation status changes to Processing.
  2. The agent classifies the prompt. If it's out of scope, the conversation ends with an explanation. See Limits and scope.
  3. For complex prompts that need three or more distinct actions, the agent emits a plan in the right-hand panel — a short list of broad steps. Simpler prompts skip planning.
  4. The agent runs one tool at a time. Each call appears in the panel as the agent decides to run it, with its inputs, status, and result.
  5. The agent posts the final answer in the transcript: a summary, key insights, and a few suggested follow-up questions.
  6. The conversation status changes to Done.

You cannot send another message in the same conversation while it's still Processing, and you can only have one Processing conversation across all your chats. Wait for the current one to finish, or switch to a different conversation.

Conversation mid-flight with a multi-step plan and pending tool calls

Inspecting evidence​

Every assistant reply links back to the tool calls that produced it.

  • Path-search results include a Query URL that opens the native Search application with the source, destination, protocol, and ports pre-filled. Use it to drill into the path or change parameters.
  • NQE results render as a sortable, copy-able table in the NQE tools panel. The NQE tool fetches at most 100 rows per query; if your question requires more, ask the agent to refine it or open the underlying query in NQE.
  • Device, interface, VLAN, VRF, host, and cloud-object lookups show structured fields directly in the plan & tools panel.
  • CVE references in vulnerability replies are clickable and carry severity.

Tracing path tool result with a clickable Query URL link in the Plan &amp; Tools panel

Following up​

After a reply finishes, the agent suggests a few follow-up questions. Click one to send it as your next prompt, or type your own. Follow-ups are scoped to the tools you have access to.

The conversation keeps the full prior context. You can reference earlier results ("for the device you mentioned in the last answer, list its interfaces") without restating them.

Renaming and deleting conversations​

In the conversations panel, right-click a conversation for Rename and Delete:

  • Rename sets a new title. Empty titles are rejected.
  • Delete removes the conversation and all its messages. There is no undo.

You can only rename or delete conversations you own.

Sending feedback​

Each assistant reply has thumbs-up and thumbs-down icons. Clicking either opens a dialog with up to 1000 characters of free-text comment. Re-submitting feedback on the same message updates your prior submission rather than creating a duplicate.

Feedback is used by the Forward AI team to improve the agent. It is not visible to other users or to your org admin.

Feedback dialog open on an assistant message

Where to go next​

  • Capabilities — the full set of question categories Forward AI can handle.
  • Limits and scope — out-of-scope replies, per-message limits, and snapshot scoping.