Guru's AI Agent Center

The AI Agent Center is your command center for managing Knowledge Agent performance and quality. It gives you visibility into what your agent is being asked, how it's answering, which sources it's drawing from, and what automated quality decisions are being made—all in one place.

👥

Access Required

Only Owners or relevant custom roles can view data regarding their Knowledge Agent in the AI Agent Center. All users can view questions they've asked in the Agent Center.

---

Monitoring your Knowledge Agent

The AI Agent Center tracks every question asked to your Knowledge Agent and the answer it provided, going back 90 days. Each entry shows the question, the answer, the sources cited, any feedback (thumbs up, flag, or none), and when it was asked and by whom.

Filtering to what matters

The filters at the top of the Questions tab are the fastest way to zero in on areas that need attention. You can filter by Knowledge Agent (if you manage multiple agents), by feedback type (flagged answers are the highest-priority review items), by date range, by asker, or by source used. You can also search for specific keywords to find questions on a particular topic.

A few filtering workflows that surface issues quickly: filter to flagged answers from the past week to catch recent problems, filter by a specific source to see whether a particular document is generating good or bad answers, or filter to "no feedback" to find questions where the answer quality is unknown.

Viewing conversation threads

When users ask questions in chat, those questions are often part of a longer conversation. Any question that belongs to a chat thread will include a View chat thread button, which opens a read-only slideout panel showing the full conversation in chronological order with the selected question highlighted.

Providing feedback

Click into any question to see the full answer and source details. From there, you can thumbs up the answer if it was accurate, flag it if it was incorrect or used outdated information, or view the sources used to generate the answer.

This feedback isn't just informational—it feeds directly into the automated verification rules that run daily, helping the system make better quality decisions at scale.

Understanding why an answer was generated

Click Answer Details on any question to see exactly how your Knowledge Agent arrived at its response. The Answer Details panel shows:

• Reason for this answer. A written explanation of how the agent constructed the answer, referencing specific citations and describing how each was used.
• Citations. The sources that were actually cited in the generated answer.
• Sources sent to the LLM. The full list of source documents sent to the model for context, including sources that were considered but ultimately not cited.
• Search terms. The keywords extracted from the user's question that were used to find relevant sources.
• Most relevant sources. The sources deemed most relevant to the extracted keywords, listed in descending priority. This is especially useful when diagnosing unexpected answers. If the right source was sent to the model but wasn't cited, the issue is likely in how the content is written. If the right source doesn't appear in the list at all, the issue is upstream—either the source isn't connected, the search terms didn't match, or the Skill's source configuration needs adjustment.

Creating a Card from an answer

Draft Card from Answer lets you turn a high-quality answer into a draft Guru Card. Only Collection Owners and relevant custom roles have access.

1. Open a trusted answer from the AI Agent Center.
2. Click Draft a Card.
3. The Card Editor opens, pre-filled with the question as the title and the answer as the body.
4. Edit and save. The Card is standalone—not tied to a specific Knowledge Agent or Collection.

Controlling what gets tracked in the Agent Center

Each Skill has an AI Agent Center tracking toggle that determines whether its invocations appear in the AI Agent Center at all. When tracking is enabled, each invocation runs post-processing to extract citations, generate an answer explanation, and classify the response as answered or unanswered. When tracking is disabled, the Skill still functions normally, but no AI Agent Center record is created.

This matters because the monitoring, filtering, and querying workflows described in this article only apply to Skills with tracking turned on. If you're not seeing a Skill's activity in the Agent Center, check whether its tracking toggle is enabled.

The default Answer Skill has tracking enabled by default. The default Research Skill has tracking disabled by default. You can enable or disable tracking on any Skill at any time under Settings > Skills on your Knowledge Agent.

---

Improving answer quality

Knowledge Agents with Skills operate differently from legacy Knowledge Agents. They aren't constrained to a simple question-and-answer format - Skills give agents specialized behavior for specific workflows, with configurable instructions and routing. That also makes them easier to train and improve, because the levers are clearer: the content the Skill draws from, and the Skill's own configuration.

Knowledge Agents with Skills

Improve the content

The most common reason for a poor answer is that the underlying source content is outdated, incomplete, or missing. Two approaches work together to keep content quality high:

Turn on quality automations. Automated verification and automated card management ensure outdated or redundant content is automatically removed or flagged before it affects answers. Auto-verify is enabled by default for all Knowledge Agents; auto-unverify can be enabled in the Quality tab when you're confident in your rules.

Query the AI Agent Center for knowledge gaps. Use chat or research mode in the AI Agent Center to identify patterns across your agent's activity - sources that are frequently cited in flagged answers, topics that consistently go unanswered, or questions where the agent is returning incomplete information. These trends point directly to content that needs to be created, updated, or removed. See Querying the AI Agent Center for trends for example prompts.

Check the Skill

If the content is accurate but the agent still isn't performing well, the issue is likely in the Skill's configuration. A Skill's description controls when the agent routes to it, and its custom instructions control how the agent responds once activated. If users are landing on the wrong Skill for their question, make the description more specific. If routing is correct but the answers aren't right, refine the custom instructions—adjust tone, specify formatting, or add guardrails for how the agent should handle certain topics. Edit Skills by going to Agent Settings > Skills on your Knowledge Agent. Learn more about configuring Skills.

Each Skill also has an AI Agent Center tracking toggle that controls whether its invocations appear in the Agent Center. Tracking must be enabled for a Skill's activity to show up in the monitoring, filtering, and querying workflows described in this article. The default Answer Skill has tracking on; the default Research Skill has tracking off. You can change this at any time under Agent Settings > Skills.

Legacy Knowledge Agents (without Skills)

For agents that don't use Skills, you can edit individual answers and assign questions to experts directly from the AI Agent Center.

Editing a suggested answer

Edit an answer by clicking into a question, selecting Edit answer, choosing the sources the answer should draw from, reviewing the regenerated response, and saving. The original asker is notified of the updated answer.

1. Click the question row to expand the details.
2. Click Answer Details, then Edit answer.
3. Under Step 1, select sources for the answer. You can search connected external sources (excluding Slack), search and apply existing Guru Cards, edit a source Card, remove a source, or create a new Guru Card (+ New Card).
4. Under Step 2, review the regenerated answer and select 🔄 refresh if needed.
5. Click Save answer to finalize.

✍️

Note

If you've added web search functionality to a Knowledge Agent, you cannot use web pages to train the AI. Guru performs a specific search at the moment the question is asked.

Assigning a question to an expert

To route a flagged or incomplete answer for expert review:

1. Click the question row to expand the details.
2. Click Assign Expert.
3. Choose an individual or Group to respond. Guru may suggest experts.
4. (Optional) Add a note for context.
5. Click Submit. The expert will be notified via their preferred notification settings. (Experts assigned via Slack will see it in My Tasks but won't receive a separate notification.)

Reviewing unverified sources

When your Knowledge Agent filters sources by verification status ("Verified only" or "Verified and no verification status"), you may see a collapsed section titled "Unverified Sources that may be relevant to this question." This surfaces sources that were relevant but excluded because they haven't been verified yet.

Expanding this section lets you review the unverified sources, open them to read full content, edit content directly (if you have verification permissions), or add a source to the answer—which automatically verifies it and triggers answer regeneration.

✍️

Note

This section only appears when your Knowledge Agent uses verification-based source filtering and relevant unverified sources exist. Users without verification permissions can view sources but cannot add them to answers.

---

Understanding your sources

The Sources tab shows which content from your connected sources is being used to answer questions—and how trusted that content is.

Each source displays a verification status: a green checkmark for verified content, a gray indicator for unverified content, or no badge for content that hasn't been reviewed yet (the default state). Depending on your Knowledge Agent's configuration, it may only use verified sources, or it may use both verified sources and sources with no verification status.

You can filter by verification status, source type, usage frequency, or Knowledge Agent.

The Sources tab is most useful for identifying high-value content (sort by usage to find your most-cited documents), finding unverified content that's being relied on heavily, and understanding coverage gaps between what's connected and what's actually being used.

---

Automated quality maintenance

The Quality tab gives you transparency into automated verification decisions. This is where your feedback from the Questions tab and the usage patterns from the Sources tab come together into automated maintenance.

How automated verification works

Automated verification runs daily, reviewing Cards and source documents that have been used in answers. Based on behavioral, content-based, and analytical rules, your Knowledge Agent automatically verifies or unverifies content. Learn more about automated verification.

Auto-verify is enabled by default for all Knowledge Agents. Auto-unverify is disabled by default—enable it when you're confident in your rules.

The Quality Log

The Quality Log is a running record of every verification event. Each entry shows which document was verified or unverified, when it happened, the reasoning behind the decision, and a confidence level:

• High confidence. Strong signals supported the decision (e.g., content thumbs-upped 10+ times with no flags).
• Medium confidence. Moderate signals (e.g., content viewed regularly but limited feedback).
• Low confidence. Limited or ambiguous signals (e.g., content used only once or twice). Worth reviewing manually.

Review the Quality Log weekly to spot unexpected changes, investigate why a source was unverified, or assess whether your verification rules need tuning. If you're seeing many low-confidence decisions, your rules may benefit from more specificity.

---

Querying the AI Agent Center for trends

Instead of manually reviewing individual questions and answers, you can use chat or research mode to conversationally query your AIAC data and surface patterns across your agent's activity.

Chat mode returns the first 50 results per request for fast, focused insights. Research mode supports deeper exploration across multiple pages of results. Both support filtering by Knowledge Agent, asker, time period, status, and source used.

Prompts to try

• "What are the top trending questions for this agent over the last 30 days?"
• "Show me all flagged questions from the past week."
• "Which questions went unanswered in the last 30 days?"
• "What topics are getting the most thumbs-down feedback?"
• "Which sources are being cited most frequently in answers that receive negative feedback?"

Regular querying helps you identify content gaps, catch Skill routing issues early, and prioritize which sources or Skills to refine next.

Pairing trends with quality automations

The insights you surface through AIAC querying become more powerful when combined with Guru's quality automations. Turning on automated verification and automated card management (auto-archive and duplicate detection) means the patterns you identify—outdated sources, content that's consistently flagged, duplicate documents competing for citations—can be addressed automatically over time rather than requiring manual cleanup. Use AIAC trends to spot the problem, then configure your automations to prevent it from recurring.

---

Frequently asked questions about the Agent Center

What's the difference between Answers analytics and the AI Agent Center? Answers Analytics (available to Admins, Knowledge Agent Owners, and relevant custom roles) shows feature adoption and usage by Group, all questions answered or not answered, and CSV export. The AI Agent Center (accessible to Experts and Knowledge Agent Owners) shows the full question, answer, sources, feedback, and actions taken. It has a 90-day history, is not exportable, and is used to review and improve AI responses directly.

Will research appear in the AI Agent Center? No. Research results can be found by the individual who conducted the research in the Chat window in the left-hand column.

What does it mean when I see a lock icon and "Answer Hidden"? This means you don't have access to the source within Guru, or the source has been deleted since the question was asked. If you have access in Guru but not in the external source (or aren't logged into it), clicking the source link will result in access being denied.

How often does automated verification run? Daily. No content is reviewed more than once within a 7-day window.

Why don't I see all verification events in the Quality Log? The Quality Log respects source permissions. If you don't have access to a document, you won't see its verification events.

What content is eligible for automated verification? Only content that has been used in an answer. This ensures verification focuses on content that actually matters.

How do I improve answers for a Knowledge Agent with Skills? Review the content your Skills reference and the Skills' own configuration. Start with the source content (is it accurate and complete?), then check the Skill's description and instructions (is it routing correctly and responding appropriately?). Learn more about configuring Skills. For additional guidance, contact [email protected].

---