Get Started with Gigamon Insights

The following section provides instructions on how to start using Gigamon Insights and using prompts effectively.

Launch Gigamon Insights

Before launching Gigamon Insights, confirm that your GigaVUE‑FM user account is assigned to the Gigamon Insights user group.

Note:  To launch and sign in to the Gigamon Insights interface, use the Gigamon Insights specific user account created for Gigamon Insights.

To start using Gigamon Insights:

1.   Enter the IP address or FQDN of the Gigamon Insights instance (configured on VMware or AWS) in a web browser. The SAML sign‑in page appears.
2. Follow the prompt to access the GigaVUE‑FM login page. Enter your GigaVUE‑FM credentials to launch Gigamon Insights.
3. Start using the Gigamon Insights chat interface.
o   The left panel provides:
•   New chat: Opens the new chat.
•   History: View prompts and responses from the past 30 days.
•   Prompts: Predefined prompts are organized into three main use‑case categories—security, network, and application use cases. Each category includes Structured, Hybrid, and Quick prompts, as described in the How to Use Prompts Successfully.
•   Create Prompt: Save personal prompts only visible to you. You can categorize these prompts.
o   Quick Actions: Curated best practice prompts based on commonly used workflows. You can select a category and subcategory to load the predefined prompts into the chat window.
o   Prompt Area: You can enter your free-form questions or use prompts from the prompt library. Additionally, you can upload a dashboard screenshot per prompt.
o   Switch LLMs: When both Amazon Bedrock and Google Vertex are configured, you can switch the active LLM between them as needed.
4. Type a prompt. You can either select a predefined prompt or type an open prompt. Review your question and any attached screenshot and submit the prompt. Gigamon Insights returns an answer.
5. Submit your feedback by selecting Thumbs-up or Thumbs-down.

How to Use Prompts Successfully

This section describes how to write effective prompts for Gigamon Insights. It focuses on three areas you control in every question:

■   Prompt and context engineering – how you describe the task.
■   Time window management – which period of data you analyze.
■   Data coverage – how much data you scan and return.

Use these practices to reduce hallucinations, make outputs explainable, and keep results repeatable.

Prompt and context engineering

Prompt and context design directly affect answer quality and the risk of hallucinations. Large language models (LLMs) balance all instructions statistically, rather than executing them like code. Overly long or conflicting prompts can still produce confident but incorrect answers.

In Gigamon Insights, you shape your questions and task prompts to get reliable, explainable results.

Use these guidelines for any question:

■   Be clear and direct. State the objective in a single sentence before adding details. For example:
“Analyze authentication failures in the past 7 days to identify suspicious login behavior.”
■   Prioritize high‑impact instructions. Use concise, high‑priority instructions instead of many low‑impact details. Put critical requirements (for example, “use read‑only queries”, “separate facts from interpretation”) early in the prompt.
■   Make uncertainty explicit. Tell the assistant how to respond when data is incomplete. For example:
“If the data is not sufficient, state that you cannot determine the answer from the available data.”
■   Avoid conflicting requirements. Do not combine instructions such as “never ask for clarification” and “must always be correct” in the same prompt. Prefer one clear priority: accuracy, speed, or minimal interaction.
■   Treat prompts as iterative. Refine your wording based on previous answers. Tighten the objective, time window, or filters until the output is consistently useful.
■   Separate facts from interpretation. Ask the assistant to structure responses into sections such as:
o   Summary
o   Evidence (directly from tools and logs)
o   Assessment (interpretation)
o   Recommendations
■   Specify output format. When you need a particular structure, say so: “Return: summary, evidence, assessment, and recommendations.” This improves consistency and makes results easier to review or export.

Prompt types and when to use them

Use one of three prompt styles depending on your task.

Table 1: Prompt Types

Prompt Type

Best for

Characteristics

Example

Structured

Complex analysis; regulated or security‑critical workflows

Longer, explicit requirements and constraints

Objective: Analyze network and authentication logs from the last 24 hours to detect port scanning, host sweeping, profiling, or spoofing.

Data scope: Use only the network flow logs and authentication logs available in Gigamon Insights. Analysis requirements: - Identify patterns that indicate scanning, sweeping, profiling, or spoofing. - Use time‑based aggregation where helpful (for example, counts per source IP per minute or hour). - Correlate authentication events with network events when needed.

Output format:

● Summary
● Evidence (including key query results and example records)
● Assessment
● Recommendations

Constraints:

● Do not infer behavior without supporting data.
● Clearly separate facts from interpretation.
● Call out any data gaps or limitations.

Quick/Short

Fast, exploratory checks; experienced users

One to two sentences; minimal structure

“Check the last 24 hours of authentication logs for suspicious activity.”

"Check the last 24 hours of network and authentication logs for port scans, host sweeps, profiling, or spoofing. Return any suspicious behavior you find."

Hybrid

Most day-to-day investigations

Brief objective plus clear output structure

“Analyze the last 24 hours of network and authentication data to detect port scanning, host sweeping, profiling, or spoofing.

Return:

● Summary
● Evidence
● Assessment
● Recommendations”

Time window management

Time window management controls which period of data the model considers. In conversation, you often use natural expressions such as “last week” or “recently”. Gigamon Insights converts those expressions into precise, inspectable time filters and can complement them with UI‑based controls.

■   Use conversational time for ad-hoc questions.
Examples: “last week”, “past 7 days”, “last 4 hours”.
■   Use precise windows for repeatable work.
For repeatable queries, investigations, and audits, specify start and end timestamps with time zones.
■   State the time range type.
o   Absolute: Specific dates (for example, “from 2025‑05‑01 to 2025‑05‑15 UTC”).
o   Relative: Rolling windows (for example, “past 7 days”).
o   Comparative: Period vs Period (for example, “compare last 7 days to the previous 7 days”).
■   Set the granularity. Specify whether you want minute-level, hour-level, or day-level analysis, depending on data volume and the kind of pattern you expect.
■   Call out business-hours or weekend rules. State whether you want to exclude weekends or focus only on business hours.
■   Reuse the prompt styles. Apply structured, quick, and hybrid prompts to express time requirements clearly and consistently.

Time expression categories and examples

Gigamon Insights is designed to handle a wide range of natural time expressions. Use the patterns below as guidance.

Absolute time range (fixed dates)

Examples:

■   “... from Jan 1 to Mar 31, 2025.”
■   “... on October 12, 2025.”
■   “... between 2025‑05‑01 and 2025‑05‑15.”
■   “... dated June 2025.”
■   “... Aug 18 between 2pm and 4pm.”
■   “... on March 3 from 9:00–11:00 UTC.”
■   “... after midnight on Jan 1.”
■   “... for Q2 2025.”
■   “... for fiscal year 2024.”
■   “... for the first week of December.”

Relative time range (rolling or anchored)

Examples:

■   “... from the last 24 hours.”
■   “... in the past 7 days.”
■   “... over the last month.”
■   “... from the last hour.”
■   “... in the last 90 days.”
■   “... since Monday.”
■   “This week’s usage.”
■   “Last month’s usage.”
■   “This quarter so far.”
■   “Week to date.”
■   “... last 30 days, excluding weekends.”
■   “Business‑hours traffic this week.”
■   “Nighttime usage over the past month.”

Comparative time range (before vs after, period vs period)

Examples:

■   “Compare this week to last week.”
■   “Month over month revenue.”
■   “Quarter over quarter growth.”
■   “Week over week error rates.”
■   “Year over year sales.”
■   “Compare last 30 days to the previous 30 days.”
■   “Compare January vs February.”
■   “7‑day average week over week.”
■   “Compare the last 14 days to the 14 days before that.”
■   “How did today compare to the same day last week?”

Mixed or domain‑specific language

Examples:

■   “Are errors worse than they were last month?”
■   “Is this week better or worse than usual?”
■   “Spikes in the last 15 minutes.”
■   “What’s changed recently?”

Data coverage

Data coverage determines how much data the system reads and returns, and whether that fits within the model’s context window and platform limits. Gigamon Insights treats coverage as an explicit, enforceable contract rather than an assumption. The guidelines are:

■   Treat each question as a coverage contract. State whether you need a sample, a bounded slice, or full coverage.
■   Use simple coverage modes. Use sample, bounded, and complete coverage and map them to clear row limits and behavior.
■   State what you need:
o   A summary or sample (small row limit).
o   A bounded view (filtered by time or attributes, often aggregated).
o   Full coverage (subject to safety limits, often for exports or audits).
■   Expect safety caps even for “full coverage”. When you ask for “full coverage” or say “show me everything”, Gigamon Insights still applies hard limits (for example, a maximum number of rows per table or per user, such as 10,000 rows).
■   Understand system behavior at high coverage. For resource‑intensive requests, the system may:
o   Return aggregated views instead of every row.
o   Ask you to narrow the time range or filters.
o   Page results and summarize each chunk rather than returning all raw rows at once.