Skip to main content
Groupe

Agentblazer Community Group

Welcome, Agentblazer! Are you ready to connect with AI leaders and enthusiasts from around the world? You've come to the right place. The Agentblazer Community is the premier destination to skill up on AI, discuss best practices, gain access to Agentforce product experts, showcase your expertise, and supercharge your career.

## Context

Building a POC Agentforce for Service on a production org (Enterprise Edition, Lightning Knowledge, Data Cloud enabled, Foundations active).

**Agent**: Employee Agent (`BotType = AgentforceEmployeeAgent`, `PlannerType = AiCopilot__ReAct`, `BotUserId = null` → runtime user = logged-in user, who is System Administrator with full Knowledge perms).

**Goal**: technician asks "How do I configure the VPN FortiClient?" → agent retrieves the relevant Knowledge article and answers.

## Problem

The action `Answer Questions with Knowledge` (`EmployeeCopilot__AnswerQuestionsWithKnowledge`, `invocationTarget = streamKnowledgeSearch`) **executes successfully but returns 0 results silently**:

```json

Output: {"knowledgeSummary": "Sorry, I can't find an answer based on the available articles."}

The same query via direct SOSL on Knowledge__kav returns the articles correctly — so it's NOT a Knowledge data problem, it's the RAG pipeline.

Event log input observed:

 

 

{

"citationsEnabled": "true",

"ragFeatureConfigId": "AiRagFeaturePromptContent_e3a977c6_c149_44c0_81fe_a6876f452fd2",

"query": "Comment configurer le VPN FortiClient ?"

}

The ragFeatureConfigId GUID is the same since the Data Library was first created — even after delete + recreate of the library, Salesforce reuses the same backend record.

What's verified OK

  • 4 Knowledge articles, PublishStatus=Online, Language=fr, IsVisibleInApp=true
  • Articles are RecordType=Procedure with content in Procedure_Steps__c (10–33k chars each, real procedure steps)
  • Data Library assigned to the agent (Knowledge type, default Data Space)
  • Content Fields include Procedure_Steps__c and all Procedure_*__c
  • Identifying Fields: Title + Summary
  • Filter by Knowledge Data Categories: OFF
  • Use Public Knowledge Articles: ON

Search Index state (via SOQL on DataSemanticSearch):

  • RuntimeStatus = READY
  • SearchType = Hybrid (semantic + keyword)
  • IndexRefreshedOn recent (post-publish)
  • 290 chunks in *_chunk__dlm, 13 contain "FortiClient" literally
  • 290 vectors in *_index__dlm
  • Max chunk length: 2104 chars (Procedure_Steps content confirmed indexed)
  • SharingStatus = REGULAR

Things I tested (none fixed it)

  1. Toggle Filter Data Categories OFF (was a forum-suggested cause)
  2. Created Data Category Group, tagged all 4 articles, configured Data Category Visibility "All" on the user profile + Default Visibility
  3. Topic language changed from en_US to fr (matches articles)
  4. Topic instructions translated to French and customized (was the generic English SvcCopilotTmpl__GeneralFAQ template)
  5. Activate/Deactivate cycle on the agent
  6. Rebuild of the Search Index (SUBMITTED → IN_PROGRESS → READY)
  7. Delete + recreate the Data Library → Salesforce reuses the same DataSemanticSearch GUID + same AiRagFeaturePromptContent_* GUID (singleton per data space? bug?)
  8. Tried REST endpoints to query the index directly in v60/v62/v66: /ssot/queries/semantic-search, /ssot/queries/hybrid-search, /ssot/retrievers → all 404

Other relevant settings

  • Knowledge.settings: enableKbStandardSharing = false (legacy sharing model)
  • Knowledge.settings: defaultLanguage = fr (matches articles)
  • Profile (System Admin): AllowViewKnowledge = true, EditKnowledge = true, ManageKnowledge = true
  • Note: Einstein Agent license users in the org have UserPermissionsKnowledgeUser blocked by license (not used at runtime here since Employee Agent uses logged-in user — but mentioning for completeness)

Questions for the community

  1. The AiRagFeaturePromptContent_* GUID — how do I inspect/debug its actual configuration? Is there a Setup UI, a Tooling API object, or a REST endpoint to retrieve its parameters (similarity threshold, top_k, default filters)?
  2. Is it expected that delete + recreate of a Data Library reuses the same backend DataSemanticSearch and AiRagFeaturePromptContent? If yes, how does one force a clean reset of the backend RAG config?
  3. Does the Data Cloud Hybrid retriever respect the legacy Knowledge sharing model (enableKbStandardSharing=false), the modern one, or its own independent permission layer?
  4. What's the correct REST path for direct hybrid/semantic search against a DataSemanticSearch index (outside the agent context, for debugging)? /ssot/queries/semantic-search 404s in API v60-v66.
  5. Has anyone else seen this exact symptom — Search Index Ready, content indexed and keyword-matching, agent returns 0 silently — and what was the fix?

Thanks for any pointers!

1 réponse
  1. Aujourd’hui, à 12:10

    @Edgar KAMDEM the chunks already contain FortiClient, so the reused GUID is not why you get zero. That canned sentence is the no-hit message. The action never got passages.

    I do not know a Setup page or a supported Tooling object that shows similarity threshold or top_k for AiRagFeaturePromptContent. The 404s on those guessed ssot paths are what I would expect. I will not invent another URL. The plan trace you already have is the supported view.

    Delete and recreate reusing the same id is ugly, and it is not this bug. The chunks are in that index. A new GUID would not change a filter that runs on the way out.

    The hybrid retriever is not SOSL. With standard sharing off, visibility is still data category, channel, and language, as the running user. A SOSL query with no language clause will return French articles that this action drops.

    Two filters SOSL skips are the ones I would check next. The user language, not the topic language. If that admin is English, French-only articles can come back empty here. And the publish channel. IsVisibleInApp is the internal app. If the library is set to public articles, look at IsVisibleInPkb. Internal procedures often are not on the public channel, and this action will not see them.

    I would stop rebuilding the index. Set that user language to French, or publish an English version, and ask once. If you still get that sentence, the channel flag is the next row.

0/9000
1 réponse
  1. Hier, à 18:58

    From your LinkedIn profile, click the plus sign. Then on the Add license or certification screen put in Name - Salesforce Agentblazer Legend '26, Issuing organization - Trailhead, and any orther pertinet information. I like to put in my trailblazer link (https://www.salesforce.com/trailblazer/mwrightms

    ) in the Credential URL spot. 

     

     

0/9000

👉 Click here to RSVP For years, my answer to is this a good architecture? was three confident words: scalable, secure, maintainable. Nobody ever asked me what those words actually meant.

 

 👉 Click here to RSVP

 

For years, my answer to is this a good architecture? was three confident words: scalable, secure, maintainable.  

 

Nobody ever asked me what those words actually meant. Including me. This Sunday, that changes. We're kicking off a 7-week live series on the new Salesforce Well-Architected Framework, and Session 1 is all about the big picture: 

 

  • ✨ What changed, and why 3 pillars became 5 
  • ✨ Why this is NOT just the AI framework 
  • ✨ A live 5-question health check you can run on your own org on Monday morning

You'll leave with a clear map of what good really looks like on Salesforce, and a language your whole team can share. Come for the pentagon. Stay for the aha moments. 📐 

 

@Salesforce Developer Group, Dubai, UAE @Agentblazer Community Group @Architect Trailblazers @* Salesforce Developers * 

0/9000
7 réponses
  1. 7 oct., 05:55

    Thanks everyone for your suggestions. 

     After completing all the necessary checks, I raised a case with Salesforce Support. 

     After reviewing the behavior and the available information, Salesforce has confirmed that this issue has been

    identified as a bug and is currently being addressed.

0/9000

Could Agentforce help improve a duplicate review process, particularly for the scenarios where existing Salesforce matching logic does not identify potential duplicates? Could Agentforce flag likely duplicates for a user to review?

2 réponses
  1. 7 oct., 04:42

    Hi,  

     

    I would recommedn asking this with copilot. In my opinion the duplicate rule should be reviewed by agentforce. It should analyze and recommend the changes to duplicate rule so it can be aligned according to your usecase. Try coworker.

0/9000

Hello. 

 

Can anyone shed some light on Einstein Request usage? As I understand it, the way to calculate the Einstein Requests used after asking and answering a question using Agentforce is to use the word count of all the prompts used in the question and answer, but I have a question here.  

 

1. By prompts used in questions and answers, do you mean the word count of the prompts that can be found in the Prompt column of the GenAIGatewayRequest with GenAIGeneration report? For reference, it took me about 5 lines of functionality to ask a question and get an answer, totaling 8 prompts. Please let me know if I should be counting the word count of all 8 prompts.  

 

2. It is my understanding that in addition to what is calculated in #1, we are charged based on the number of records processed. Is this the part that is used to calculate the Data Cloud Credit? I would be grateful if you could provide the formula for the number of records to be charged.  

 

 

1 réponse
  1. 6 oct., 09:06

    Disclosure: I work at Ashapura Softech, a Salesforce implementation partner. I can only answer part 1 with confidence. 

     

    On question 1: Salesforce's current published rate card does not meter prompts by word count. Prompt usage is billed in 2,000-token chunks, rounded up, and the prompt tier (Basic, Standard or Advanced) sets the credits per chunk. So the unit is tokens, not words, and a short prompt still costs one full chunk; counting words in the Prompt column will only give a rough estimate. 

     

    On whether to count all 8 prompts: in the GenAIGatewayRequest report each row is a separate model call. My reading is that the rounding up applies per request, so I would count the chunks for each of the 8 requests and add them up, instead of combining the text first. Please treat that as my interpretation, not something I have seen stated in the rate card. 

     

    On question 2: I could not find a published formula for records processed and Data Cloud credits, so I would rather not guess. Your Salesforce account executive or the billable usage documentation in Salesforce Help is the safest source. 

     

    One caution: your thread is from February 2025 and Salesforce has changed Agentforce pricing since then (the current rate card took effect in April 2026), so check which pricing applies to your own order form. Digital Wallet in your org shows which usage type is consuming credits.

0/9000

🚀 Apex Log MCP 2.0.1 for Salesforce — AI-Powered Debug Log Analysis!

Debugging Salesforce Apex logs can be time-consuming, especially when you're dealing with large and complex logs.

With Apex Log MCP 2.0.1, you can bring MCP and AI into your Salesforce debugging workflow. 🤖

In my latest video, I cover:

✅ What is Apex Log MCP? 

✅ How to set it up and configure it 

✅ Connect it with an AI assistant using MCP 

✅ Analyze Salesforce Apex debug logs with AI 

✅ Identify errors and performance issues faster 

✅ Practical demo and real-world usage

This is a great example of how MCP + AI can improve the Salesforce developer experience.

🎥 Watch the full video:

 

https://www.youtube.com/watch?v=QUdbooXcqgs

 

📺 Subscribe to Salesforce Hunt:

 https://www.youtube.com/@SalesforceHunt

 

@Agentblazer Community Group @* Sales Cloud - Getting Started *@* Release Readiness Trailblazers *@* Sales Cloud - Best Practices *@* Salesforce Administrators *@* Service Cloud * @* Salesforce Administrators *@* Salesforce Platform *@* Service Cloud *@* Sales Cloud - Getting Started *@* Sales Cloud - Best Practices *@Salesforce Flow Automation

 

🚀 Apex Log MCP 2.0.1 for Salesforce — AI-Powered Debug Log Analysis!Debugging Salesforce Apex logs can be time-consuming, especially when you're dealing with large and complex logs.

2 commentaires
  1. 5 oct., 12:44

     [Free & Updated] Salesforce Exam Practice Questions

     

    Looking for real & accurate questions and answers? 

     

    Try the free demo on: QA4Exam 

0/9000

Is it possible to ask an agent to draft an email using a specific email template? I would like to be able to ask my agent something like, "Draft an email response for this case using this "My Template Name" email template. We have several hundred email templates used across our org by the many different Teams.

4 réponses
  1. 29 sept., 05:21

    Hi @James Waltz

     

    Yes, but I couldn't confirm it works out of the box. The reliable route is a small custom agent action. 

    What I verified 

    • Employee-facing agents have a standard Draft or Revise Email action, which the reasoning engine picks automatically for drafting emails. It generates text with the LLM. I couldn't confirm it can take a template name and merge that template's content. The Help page for that action didn't load for me, so check it in your org.
    • The Agentforce Service Agent on Email feature does use a Lightning email template, but only as a single wrapper structure for autonomous replies. It isn't a "pick any template on request" mechanism, so it doesn't fit your use case.

     

    What I'd recommend (my design, not a documented feature)

     

      

    Build a custom agent action, a Flow first, with Apex only if needed: 

     

    1. Inputs: template name and Case ID, collected from the rep's prompt.
    2. Lookup: query EmailTemplate by name. With hundreds of templates, return the top few matches when the name is ambiguous, and let the agent ask the rep to pick one.
    3. Render: merge the template against the Case, using Apex Messaging.renderStoredEmailTemplate or an equivalent Flow approach.
    4. Return: hand the rendered body back to the agent. Optionally pass it through Draft or Revise Email to tailor the tone to the case.
    5. Draft only: never send from this action. Have the rep review and send.

     

    Things to check before building

     

     

    • Access: confirm the action respects template folder sharing, so reps only get templates they can already see. This matters with many teams' templates in one org. I haven't verified how the agent's running user affects this, so test it.
    • Template type: Lightning and Classic templates use different merge syntax ({{{Case.Subject}}} vs {!Case.Subject}), so confirm which types you have.
    • Naming: with this many templates, exact-name matching will be brittle. A naming convention or a metadata field such as team or use case will help.

    I hope you find the above information helpful. If it does, please mark it as Best Answer to help others too.

0/9000

🚀 Next Week: One workshop. Two chances to get hands-on with AIforce.

 

Join us next week for Get Hands-on with AIforce, a free virtual workshop where you’ll build a real Salesforce app using Agentforce Vibes, the Headless Experience Layer, APIs, MCP tools, Skills, and more. Learn how to build composable, headless, and agent-ready experiences that extend Salesforce beyond the platform — and put those concepts into practice. 

 

Register for the session that works for you and come ready to build: 

🔹 October 7 | 10:30 a.m. SGT / 1:30 p.m. AEDT ➡️ https://sforce.co/4rxnHkx

🔹 October 7 | 10:00 a.m. EDT ➡️ https://sforce.co/4d3Ov6c

🚀 Next Week: One workshop. Two chances to get hands-on with AIforce.

5 commentaires
0/9000

Salesforce just reimagined its Well-Architected Framework, and @Salesforce Developer Group, Dubai, UAE is launching a brand new 7-part series to help you put it to work. From trust and security (for people and autonomous agents) to resilience, DevOps, cost optimization, and ethical, human-centered design, this series is a structured, practical deep dive into what makes a Salesforce solution truly well-architected.  

 

11 Oct 2026 - Session-1: The Blueprint Unveiled 

https://trailblazercommunitygroups.com/j/sxgzc6bvd9k5h/

 

18 Oct 2026 - Session-2: Trust, Earned Not Assumed 

https://trailblazercommunitygroups.com/j/ntzrw63pjpn94/

 

25 Oct 2026 - Session-3: Built to Bend, Not Break 

https://trailblazercommunitygroups.com/j/rsnter9h3mhnb/

 

1 Nov 2026 - Session-4: Running It Like You Mean It 

https://trailblazercommunitygroups.com/j/g9zpny63ddc7w/

 

8 Nov 2026 - Session-5: Every Credit Counts 

https://trailblazercommunitygroups.com/j/2a85jc499dbvv/

 

15 Nov 2026 - Session-6: Fair by Design  

https://trailblazercommunitygroups.com/j/5g8hdwy6jmwp5/

 

22 Nov 2026 - Session-7: Inside the Review Room 

https://trailblazercommunitygroups.com/j/bzx6y8tt8bkxy/

 

@Agentblazer Community Group @Architect Trailblazers @* Salesforce Developers * @WFD Partner Cohort: TDX2630DaysChallenge

0/9000