AI Agent Platform, Discovery & Configuration Playbook - Support and Troubleshooting

AI Agent Platform, Discovery & Configuration Playbook Summary

A comprehensive troubleshooting playbook covering Now Assist Panel rendering failures, AI Agent and skill discovery pipeline breakdowns, plugin version management after upgrades and clones, assistant configuration, skill cache performance optimization, subscription consumption anomalies, and platform-level UI issues including the agentic support toggle and common NAP trigger patterns.

📂 Category: AI Agent — Platform & Configuration ⚡ Applies to: NAP · AI Agent Studio · NAVA · Conversational Studio · Skill Kit · Triggers Use Cases in This Playbook 🗺️ Flow Map — How a NAP Request Is Resolved 🧭 Decision Tree — Which Fix Do I Need? 1 NAP Renders Blank — Duplicate Assets & Channel Flags 2 Agent Invisible in NAP — Discovery Pipeline Breakdown 3 Post-Upgrade/Clone Catastrophe — Plugin Version Mismatch 4 NAP Toggle Silently Reverts — Missing Skill Backing Records 5 Ghost Agents — Inactive/Unauthorized Agents in Search Results 6 Deprecated Agent Trap — "No Specialized Agents Available" 7 20-60 Second Delay — Skill Cache Overflow & BGE Reranker Lag 8 CI Summarization Missing from Conversational Panel 9 Assistants Page Dead — Unified Developer Shell Only 10 "Engage via VA Assistants" Option Vanishes 11 Runaway Consumption — Automated Triggers Draining Assists 12 Phantom Usage — Empty User & Skill Config in Logs 13 Multi-Language Meltdown — Non-English Agent DataType Errors 14 Agentic Support Toggle Won't Stick in VA Assistant 15 Seven Deadly Patterns — Common NAP & Trigger Failures 16 Studio Skill-Config Binding Mismatch — config-id=undefined 17 Custom Agents Hidden — Fulfiller Entitlement & AI-Native Tier 18 Studio Test Button Greyed & Define-User-Access 1000-Row Cap 19 Agentic Orchestration Auto-Flips ON After Plugin Upgrade ⚙ Key Plugins & Properties Reference 📚 Consolidated Reference Links 🗺️ Flow Map — How a NAP Request Is Resolved Every Now Assist Panel request flows through the seven stages below. Where the chain breaks tells you which Use Case to open. Use this map either as a training overview of the platform or as a "where am I in the chain?" diagnostic compass before opening a specific UC.

Map 1 · Platform & Discovery Lifecycle (high-level) 1 NAP UI renders & loads → 2 Plugins versions in sync → 3 Discovery Pipeline 7-table chain → 4 Studio Binding config-id matched → 5 Agentic Toggle orchestration enabled → 6 Performance & I18N cache + reranker + lang → 7 Consumption assists logged Stage details — which Use Case lives at each stage Stage 1 · NAP UI Layer The panel must render and the Assistants page must load before anything downstream can be reached UC 1 UC 9 Stage 2 · Plugins & Entitlement Plugin versions must be aligned; entitlements (fulfiller / AI-Native tier) must be present after upgrade or clone UC 3 UC 17 Stage 3 · Skill Discovery Pipeline The 7-table chain (see Map 2 below) must resolve the user's utterance to a discoverable agent / skill UC 2 UC 4 UC 5 UC 6 UC 8 UC 10 Stage 4 · AI Agent Studio (build & test) Studio configuration — skill-config binding, ACL test access, role dropdowns — must be intact before activation UC 16 UC 18 Stage 5 · Agentic Orchestration Toggle The Agentic Support toggle must hold its intended state across saves, plugin upgrades and clones UC 14 UC 19 Stage 6 · Performance & Internationalisation SkillDetailsCache, BGE reranker timing, and multi-language behaviour all gate response speed and correctness UC 7 UC 13 Stage 7 · Consumption & Reporting Once a request executes, assists must be logged correctly and not double-counted by runaway triggers UC 11 UC 12 Quick reference Seven recurring quick-check patterns (proficiency empty, fulfiller can't see NAP, trigger not firing, etc.) UC 15 Map 2 · Skill Discovery Pipeline — the 7-table chain (Stage 3 expanded) When a user types in NAP, this chain must resolve fully and in order. A break at any point means "agent invisible / no agents found" — open UC 2 for the diagnostic walkthrough.

① CONTEXT PROFILE sys_cs_context_profile ⚠️ active=true required → ② PROFILE↔SKILL sys_cs_context_profile_skill m2m link → ③ SKILL DEFINITION sys_gen_ai_skill skill_document populated → ④ AIA METADATA sn_aia_skill_metadata paired with ③ ↓ ⑤ APPLICABILITY sys_gen_ai_skill_applicability table/scope match → ⑥ DEPLOYMENT CONFIG sys_now_assist_deployment_config_attributes per-deployment attrs → ⑦ FILTER sys_gen_ai_filter final eligibility gate → ✓ AGENT DISCOVERED visible in NAP ready to invoke ℹ️ How to read Map 2: If the agent is not appearing in NAP, walk this chain from ① to ⑦ in UC 2 . The first table whose record is missing, inactive, or out of sync with the others is the break point. Common cluster: ① inactive after clone (PRB1852997), ③ + ④ orphaned skill records (also drives UC 7 cache-overflow latency), ⑥ stale deployment-config record after upgrade.

🧭 Bridge to the Decision Tree: If you already know your symptom, jump to the decision tree below — it routes symptoms directly to the relevant UC. Use the flow map above when you need the architectural context first; use the decision tree when you have a specific failure signal.

🧭 Decision Tree — Which Fix Do I Need? Does the Now Assist Panel open at all? No — blank screen or spinner → Use Case 1 (check duplicate assets, channel flag, Zurich P1 regression) Yes, but agent not found → Continue Did this issue start after an upgrade or clone? Yes → Use Case 3 (follow comprehensive post-clone checklist) No → Continue Does the agent work in Studio but not in NAP? Yes → Use Case 2 (check 7 root causes in order) NAP toggle won't enable → Use Case 4 Agentic Support toggle reverts → Use Case 14 Is it a common quick-check pattern? Fulfiller can't see NAP / trigger not firing / proficiency empty → Use Case 15 (Seven Patterns A-G) "No specialized agents available"? Yes → Use Case 6 (Deprecated Agent — migrate to replacement) Is agent invocation very slow (20+ seconds)? Yes → Use Case 7 (SkillDetailsCache + orphaned skill cleanup + BGE Reranker) Is the Conversational Studio / Assistants page blank? Yes → Use Case 9 (Upgrade sn_oe_sfs) Is it a consumption or reporting issue? 37,000+ actions / runaway triggers → Use Case 11 Dashboard shows 0 / empty user fields → Use Case 12 Are agents failing in non-English languages or wrong dialect? DataType failure (especially Japanese) → Use Case 13 / Sub-A (Upgrade sn_aia to 6.0+) BR-PT vs PT-PT or other dialect drift → Use Case 13 / Sub-B Citation strings remain in English → Use Case 13 / Sub-C (PRB2019099) Studio shows "ACLs not defined" but standalone skill-kit URL has them? Yes — URL contains config-id=undefined → Use Case 16 (rebind sn_aia_agent_tool_m2m to active skill-config) Custom agents missing in Studio create-and-manage page? "Unable to locate data broker now_assist_for_platform_active" → Use Case 17 (run hasLicenseForArtifact entitlement check; install AI-Native tier plugin) Test button greyed out OR Define-User-Access role dropdown cut off? Yes → Use Case 18 (PRB1978409 sibling roles-dropdown cap; edit sys_agent_access_role_configuration via list) After plugin upgrade, Agentic Support flipped ON unexpectedly? Toggle was OFF, now ON → Use Case 19 (set agentic_orchestration_enabled=false post-upgrade; KB2913453) Toggle won't enable → Use Case 14 🖥️ Use Case 1: NAP Renders Blank — Duplicate Assets, Channel Flags & Zurich P1 Regression NAP Plugin Problem The Now Assist Panel fails to render — showing a blank white screen, spinner, or only the application shell when clicking the Now Assist icon. No prompt interface appears. The issue typically surfaces after a platform upgrade, instance clone, or plugin repair.

Symptoms Blank white screen or empty shell when opening the Now Assist icon in the header. Perpetual loading spinner — the input prompt never renders. Browser console shows errors referencing duplicate asset files, script bundles, or 404s on UX framework resources. Issue appears immediately after a platform upgrade, instance clone, or plugin repair. NAP hangs after upgrade to Zurich Patch 1 specifically. Root Causes Duplicate asset file exists in the sys_ux_lib_source_script table, causing the UX framework to fail during rendering (PRB1887044). Plugin version mismatch: Family-scoped plugins were upgraded but store applications were not. NASDE application ( sn_vad_genai ) outdated and incompatible with the current platform version. Stale HTTP connection endpoints after an instance clone. sys_cs_channel "supports_multiple_conversations" flag set to wrong value on the Now Assist Panel channel — when enabled, can cause panel not to load (PRB1965617). Zurich Patch 1 regression: AI Engagement Experience plugin needs upgrade to v1.0.6. Workaround: enable Voice Input. Resolution 1 Check sys_ux_lib_source_script for duplicate asset files. Identify and delete the duplicate entry per PRB1887044. 2 Check sys_cs_channel → Now Assist Panel record → supports_multiple_conversations value. If set incorrectly, restore to OOB value (PRB1965617). 3 For Zurich P1 blank screen specifically: upgrade AI Engagement Experience plugin to v1.0.6 OR enable Voice Input as a workaround. 4 Repair all Now Assist plugins in order: Now Assist Core → Generative AI Controller → Platform AI Agents and Skills → Now Assist Platform Skills → Now Assist Admin Console . 5 Upgrade all Now Assist store applications to the latest compatible version via App Manager. 6 After instance clone: re-repair all plugins and re-verify HTTP connection endpoints for llmproxy-prod-gateway and mlprediction-prod services. 7 Clear cache ( cache.do ) and test in a new conversation. ℹ️ Plugin Repair Order is Critical: Always repair in the exact sequence above. The repair order reflects dependency chains — repairing out of order can leave stale configurations in place. After clone, UPDATE all plugins first, then REPAIR any that still show issues.

📘 KB2507741 — Plugin Repair Sequence 🔴 PRB1887044 — Duplicate asset file 🔴 PRB1965617 — supports_multiple_conversations flag 🔍 Use Case 2: Agent Invisible in NAP — Discovery Pipeline Breakdown & Context Profile De-Sync Discovery NAP Config Problem AI Agents or Agentic Workflows that function correctly in the AI Agent Studio playground do not appear as options or tile buttons in the Now Assist Panel. NAP responds with generic search-based answers or routes to "Get Help" instead of invoking the intended agent. The issue may exist in PROD but not STG/DEV on the same patch version. This is one of the most complex diagnostic scenarios because the skill discovery pipeline has 7+ tables that must all be in sync.

Symptoms Agent works in AI Agent Studio playground but is not triggered from NAP with the same utterance. NAP routes to "Get Help" or returns generic answers instead of invoking the agent. Workflow tile missing from the NAP skill list despite being active. sys_cs_skill_discovery_tracking shows no records or empty "Available Skill IDs" for the conversation. All conversation tasks are suspended in the conversation record. Works in STG/DEV but not PROD — or works for one user but not another. Root Causes (in order of frequency) #1 Context Profile Inactive: sys_cs_context_profile for the assistant is set to active = false , de-syncing the deployment record. A Business Rule on sys_now_assist_deployment syncs state — manual profile changes break this synchronization. This is the most common root cause. #2 Not Promoted: Skills, agents, and topics were not promoted for the corresponding assistant. Items must be promoted in Conversational Studio to show as a pill button in NAP. #3 Missing sys_gen_ai_skill: The sys_gen_ai_skill record is absent while sn_aia_skill_metadata exists — the UI assumes all records are intact and skips recreation. #4 NAP Not Enabled: Agent not enabled for NAP in the "Select UI Display" step of the workflow guided setup. #5 Missing User Role: Some OOB agents require specific roles (e.g., sn_uxc_gen_ai.platform_ai_analyze_trnds for Analyze Task Trends). User missing the role sees "no specialized agents currently available." #6 Agent Proficiency Empty: The proficiency field on sn_aia_agent is empty — platform cannot find any agents and returns "Unable to find any agents to complete the mission." #7 Plugin Version Mismatch: Plugins installed in PROD differ from DEV; missing plugin causes discovery failure. Diagnostic Steps 1 Check skill discovery tracking: Query sys_cs_skill_discovery_tracking filtered by conversation.sys_id . If "available_skills" is empty → discovery pipeline is broken. Script: var gr = new GlideRecord('sys_cs_skill_discovery_tracking'); gr.addQuery('conversation.sys_id', ' '); gr.query; while(gr.next) { gs.info('Available Skills: ' + gr.getValue('available_skills')); } 2 Check context profile: Query sys_cs_context_profile for the assistant → verify active = true . Check update history for unauthorized changes. 3 Check agent proficiency: Open sn_aia_agent → verify proficiency field is NOT empty. 4 Compare plugin versions: Run Discovery_diagnostics.txt script between working and non-working instances. Pay attention to sn_vad_genai , sn_aia , sn_ais_assist versions. 5 Check full table chain: sys_cs_context_profile → sys_cs_context_profile_skill → sys_gen_ai_skill → sn_aia_skill_metadata → sys_gen_ai_skill_applicability → sys_now_assist_deployment_config_attributes → sys_gen_ai_filter . 6 Check user roles: Verify user has required role for the specific agent (e.g., now_assist_panel_user for NAP access). Resolution by Root Cause 1 Context Profile Inactive: Reactivate sys_cs_context_profile , OR deactivate/reactivate the assistant via the Assistant Setup page to restore deployment sync. Never directly modify the context profile record. 2 Not Promoted: Promote all skills, agents, topics, and workflows for the assistant in Conversational Studio. 3 Missing sys_gen_ai_skill: Manually recreate it to match the sn_aia_skill_metadata entry. 4 NAP Not Enabled: Enable in workflow guided setup → "Select UI Display." Save and confirm it persists. 5 Missing User Role: Add the required role (e.g., sn_uxc_gen_ai.platform_ai_analyze_trnds ). 6 Empty Proficiency: Set proficiency on the sn_aia_agent record. 7 After ALL changes: reindex the Skills indexed source. 💡 Key Insight: There is a Business Rule on sys_now_assist_deployment that syncs state with sys_cs_context_profile . Directly modifying the context profile record bypasses the assistant UI and can desynchronize them. Always toggle assistant state via the Assistant Setup page UI , never by direct record modification. When comparing instances, compare STG with PROD (not DEV with PROD) as DEV may have test configurations.

📘 KB2302120 — Workflow Not Appearing in NAP 📘 KB2541917 — Plugin Version Check Script 🔄 Use Case 3: Post-Upgrade/Clone Catastrophe — Plugin Drift & Stale Script Includes Plugin Config Problem After a platform upgrade, instance clone, or patch application, Now Assist features break due to version incompatibilities between family-scoped plugins and store applications. This is the single most common root cause across all AI Agent issue categories.

Symptoms Assistants page shows only the Unified Developer App Shell — no data loads. Tools section empty for agents that have tools assigned (especially RAG Retrieval tools). Features that worked before the upgrade fail silently with no errors in logs. sys_upgrade_history table missing upgrade records for certain store apps. Customer-modified script includes blocked the upgrade from overriding them. "Cannot invoke getApiKey because apiKeyCredential is null" after clone. Skills show blank LLM Service field in sn_nowassist_skill_config . Now Assist Panel deployment disabled after clone. Semantic Index Configuration disabled after clone. Root Causes Family-scoped plugins upgrade automatically; store applications require manual upgrade — they are NOT automatically updated. AI Search RAG ( sn_ais_rag ) on version 4.0.x is incompatible with newer release — must be 5.0.2+. After cloning, sys_gen_ai_feature_mapping records may be deleted — the AIA ReAct Engine record goes missing. NowLLM requires this. Omni-Experience Standard Feature Set ( sn_oe_sfs ) outdated, blocking Conversational Studio. sys_alias table missing "LLM Proxy OEM" record after clone (sys_id: e19c3e219fabd210da502fca9a0a1cab ). Customized script includes (e.g., AgentExecutionPlanUtil , AIAConversationUtilsAjax ) blocking upgrades. Comprehensive Post-Clone/Upgrade Checklist 1 UPDATE all plugins to latest compatible versions via App Manager. Key plugins: sn_vad_genai , sn_ais_rag , sn_generative_ai , sn_aia , sn_oe_sfs . 2 REPAIR plugins that still show issues after update (in dependency order). 3 Verify LLM Proxy OEM exists in sys_alias table. If missing, repair Generative AI Controller. 4 Enable Now Assist Panel deployment in sys_now_assist_deployment if disabled. 5 Activate Semantic Index Configuration in ais_semantic_index_configuration if disabled. 6 Check sys_connection for Azure OpenAI HTTP connection records. 7 Check sys_gen_ai_feature_mapping for AIA ReAct Engine record. If missing, remove stale update sets and repair sn_aia . 8 Revert customized script includes to OOB versions — check sys_upgrade_history for customizations that weren't reverted. 9 Verify sn_nowassist_skill_config skill fields are populated (not empty/null). 10 Test a simple Now Assist skill to confirm end-to-end connectivity. ⚠️ Critical: Always check sys_upgrade_history after any upgrade to verify store app updates were actually applied. Version numbers in the App Manager can be misleading if the actual update content was not installed. Never install Generative AI Controller standalone — install via Now Assist suite applications.

replace_on_upgrade Mechanics — Field-Verified Detail Two distinct mechanisms produce the same family of post-upgrade symptoms in production environments:

(a) Customizations persisted with replace_on_upgrade=false : Changes written into a plugin scope ( sn_aia , sn_uxc_gen_ai , sn_vad_genai , sn_generative_ai ) without flipping replace_on_upgrade=true . At plugin upgrade or repair, the upgrade engine SKIPS those records, leaving stale code in place. Two instances thought to be on the same plugin version then diverge in behaviour. (b) Family-version drift: The instance is on a Glide family (Yokohama / Zurich / Australia) whose latest plugin is several minor releases behind the issue's fix. Either the fix is gated behind a higher store-app version that requires a Glide patch upgrade, or the store app is installed but its prerequisites (e.g., AIA 7.x for Skill-Kit 8.x) are not present. When prerequisite versions are not aligned, agentic evaluation may fall back to executing as a guest user. Skipped-Records Diagnostic (run after every upgrade or repair) 1 Find skipped records for the most recent upgrade across AIA scopes — query sys_upgrade_history_log_list.do filtered by upgrade_history.to_versionIN sn_aia,sn_uxc_gen_ai,sn_generative_ai AND dispositionIN 4,9,10,11 (these dispositions mean the record was skipped because of replace_on_upgrade=false or because a copy already existed). 2 Find AIA-scope records that are now blocking upgrades — query sys_update_xml_list.do filtered by the AIA application sys_id ( 32808450431302106c3603295bb8f245 for sn_aia ), grouped by type . Records with replace_on_upgrade=false are the candidates. 3 For each skipped record: set replace_on_upgrade=true , then repair the plugin. Repair uses the same evaluator as upgrade — repair alone, without first flipping the flag, leaves the skipped records skipped. 4 "Revert to Base System" only works on records that exist but were customized. "Revert copy to Base System" is the only path for records that have a customer copy but were never installed in the current scope. 5 Validate on a sub-prod copy first. Clone-and-test before applying the fix to production is the recommended approach. Mandatory Verification Scripts (post-upgrade) 1 License check — var ru = new sn_aia.AiAgentRuntimeUtil(); gs.info(ru.hasValidLicense()); (expect true if not an entitlement gap). 2 Fulfiller entitlement check — gs.info(new sn_entitlement.GlideEntitlementAPI().hasLicenseForArtifact("family_id","sn_gen_ai_platform_fulfiller",false,true)); ( false → entitlement gap; route to UC16 ). 3 Display-name sanity check on AIA tables — new GlideRecord('sn_aia_property').getDisplayName() , new GlideRecord('sn_aia_agent_tool_m2m').getDisplayName() . Both should return "name" ; PRB2001660 returned a wrong field and is fixed in AIA 7.1. Recommended Actions and Common Pitfalls Action Outcome Sub-prod clone → upgrade → confirm → push to prod ✅ Effective approach for plugin-drift scenarios "Revert to base system" on a small set of skipped Platform AI Agents and Skills files ✅ Restores summarize functionality Marking specific files as "Replace on upgrade" before plugin update ✅ Prevents the skip on the next upgrade Repairing the plugin without first flipping replace_on_upgrade=true ❌ Skipped records remain skipped — common anti-pattern Marking replace_on_upgrade=true only on sn_aia when the actual gap is a missing sn_vad_genai module path ❌ No change — the actual fix path is the sn_vad_genai module, not sn_aia ✓ Pre-upgrade gate (recommended): Any sys_update_xml record in an AIA scope with replace_on_upgrade=false should block the upgrade until reviewed. Customer customizations should set replace_on_upgrade=true OR live in a customer scope, never in the OOB plugin scope.

🔴 PRB1968410 — Store app not updated 🔴 PRB2010061 — AI worker execution fails from BR (sn_aia_execution_plan runs as guest); fix in AIA 8.0 🔴 PRB1999299 — Customised UXC files unchanged after upgrade (SIR Resolution Notes regression) 🔴 PRB2001660 — AIA "AI Agent property display field is name"; fixed in Now Assist AI Agents v7.1 🔴 PRB1972902 — Deflection state recorded incorrectly post-topic execution 📘 KB2655821 — Related symptoms 📘 KB2541917 — Plugin Version Check Script 📘 KB2666962 — Pre-Validated Suite Versions 📘 KB2949962 — Known error for PRB2001660 🔒 Use Case 4: NAP Toggle Silently Reverts — Missing Skill Backing Records NAP Config Problem Enabling the Now Assist Panel for a workflow fails silently — the NAP toggle in "Select UI Display" reverts to disabled after saving, or the workflow shows as read-only and cannot be activated.

Root Cause The sys_gen_ai_skill record is missing while the sn_aia_skill_metadata record exists. The UI logic assumes all backing records are intact and does not recreate the missing entry — the "active" flag never flips correctly.

Resolution 1 Manually recreate the missing sys_gen_ai_skill record to match the existing sn_aia_skill_metadata entry. 2 Alternatively, deactivate and reactivate the assistant via the Assistant Setup page to trigger automatic recreation of all backing records. 3 Never manually delete or modify sn_aia_skill_metadata or sys_gen_ai_skill records — they are critical for skill discovery and NAP activation. 👁️ Use Case 5: Ghost Agents — Inactive or Unauthorized Agents Haunt NAP Search Results Discovery NAP Problem AI Search returns inactive or unauthorized agents during skill discovery, causing users with restricted roles (e.g., snc_external ) to see agents they cannot access. Selecting these agents produces "access denied" errors. Deactivated agents continue appearing as options in VA portals.

Root Cause The _getSkillToAgentMap function under Multi Content Response uses a plain GlideRecord call (not GlideRecordSecure) when mapping skills to agents. This ignores ACL checks entirely. ACL enforcement only occurs when the user attempts to access the agent — not when it's suggested . This creates a gap: agents are "suggested" but can't be "accessed."

Resolution 1 Ensure agents intended to be hidden are fully deactivated AND their associated skills are also deactivated. 2 Remove the agent from the assistant's skill mapping entirely for immediate relief. 3 Monitor for platform fixes introducing ACL-aware agent discovery in the Skill-to-Agent mapping. ℹ️ Platform Gap: This is a known architectural gap. The fix requires changes to Multi Content Response → _getSkillToAgentMap to use ACL-checked GlideRecord calls instead of plain queries. ACL enforcement during skill-to-agent mapping is being reviewed for future enhancement.

⚠️ Use Case 6: Deprecated Agent Trap — "No Specialized Agents Available" & Missing Roles Discovery Config Problem Users receive "There are no specialized agents currently available" messages, or the wrong agent is invoked, because the requested workflow has been deprecated and replaced.

Deprecated → Replacement Mapping Deprecated Replacement Notes "Analyze Incident Trends" "Analyze Task Trends" Requires role: sn_uxc_gen_ai.platform_ai_analyze_trnds . If user sees "organize_general_knowledge" tool missing → using deprecated version. "Check Ticket & Support Status" topic Request Status AI Agent KB2641951. Old topic has date handling issues. Resolution 1 Deactivate or remove deprecated agents/topics to prevent conflicts during skill discovery. 2 Upgrade the corresponding AI Agent collection plugins to the latest version to get replacement agents. 3 Ensure users have the required role for the replacement agent. 4 For "Analyze Task Trends": requires GAF ITSM Grouping skill (NOT "GAF Suggested Steps ITSM Grouping" — that's for Next Best Action), ~2,000+ records for meaningful clustering, and the "Issue trend analysis AI agent" must be active. 📘 KB2641951 — Request Status AI Agent Upgrade ⚡ Use Case 7: 20-60 Second Delay — Skill Cache Overflow & BGE Reranker Lag Performance NAP NAVA Problem AI Agent invocation from NAVA or NAP takes 20–60+ seconds to start, even though the same agent loads instantly in AI Agent Studio. Gen AI logs show "Processing for feature invocation … timed out after 60000 milliseconds." BGE reranker async calls take 17+ seconds on some instances.

Root Causes SkillDetailsCache overflow: Default cache size is too small for the number of skills, causing repeated cache misses and slow discovery on every invocation. Orphaned sys_gen_ai_skill records: Records with empty skill_document field (caused by deleting agents outside AI Agent Studio) exceed the default cache size of 100. BGE Reranker latency: The reranker step during discovery takes 17+ seconds on async calls — a known performance limitation. Resolution 1 Increase SkillDetailsCache: Set glide.cache.size.SkillDetailsCache to accommodate the number of skills (e.g., 450+ for large instances). 2 Clean orphaned skill records: Delete records in sys_gen_ai_skill where skill_document is NULL. Script: var gr = new GlideRecord('sys_gen_ai_skill'); gr.addQuery('skill_document', ''); gr.query; gs.info('Orphaned: ' + gr.getRowCount); 3 Deactivate unused skills and agents to reduce the discovery workload. 4 Monitor "AIA-Agent Invoker" latency in Gen AI logs to measure improvement after cache adjustments. 📘 KB2558796 📘 KB2446818 🔴 PRB1974150 — NAVA discoverability ⚙️ glide.cache.size.SkillDetailsCache 📋 Use Case 8: CI Summarization Missing from Conversational Panel NAP Config Problem The CI Summarization CMDB skill works when using the Summarize button on a Configuration Item record page, but is not available within the Now Assist Panel conversational interface.

Root Cause & Resolution CI Summarization CMDB was designed to work only on record page context , not within the conversational NAP interface. The skill configuration does not have a Panel display option.

1 Confirm with product documentation that this skill is currently supported only on record pages, not in NAP. 2 For panel-based summarization, use the generic "Summarize a record" skill, or build a custom skill using Now Assist Skill Kit with App Engine. 📱 Use Case 9: Assistants Page Dead — Unified Developer Shell Only After Multi-Version Upgrade Plugin Config Problem Navigating to Conversational Interfaces → Assistants shows only the Unified Developer App Shell without loading any data. Common after multi-version upgrades (e.g., Xanadu → Yokohama → Zurich).

Resolution 1 Upgrade the Omni-Experience Standard Feature Set ( sn_oe_sfs ) plugin to the latest version compatible with the platform release. 2 Repair the Now Assist Admin Console ( sn_nowassist_admin ) plugin. 3 If the instance was upgraded through multiple versions, verify ALL dependent plugins are on compatible versions. 🔴 PRB1968410 📘 KB2655821 🔗 Use Case 10: "Engage via Virtual Agent Assistants" Option Vanishes Config NAVA Problem The "Engage via Virtual Agent Assistants" option is missing from AI Agent Studio for certain agents, preventing them from being linked to VA assistants.

Resolution 1 Compare the sys_now_assist_deployment_config and sys_now_assist_deployment_config_attributes records between a working OOB instance and the affected instance. 2 Verify sys_now_assist_deployment and related configuration attribute records are present and correctly linked. 3 Repair the Now Assist AI Agents plugin ( sn_aia ) and verify the UI option reappears. 💳 Use Case 11: Runaway Consumption — Automated Triggers Silently Draining 37,000+ Assists Config Trigger Problem Now Assist subscription assists are consumed unexpectedly — 37,000+ actions in a single day — by automated triggers (e.g., itsm.aia.worker ) running workflows on every incident creation. Each AI Agent execution consumes 25 assists per invocation . Users with now_assist_panel_user role can trigger agents, and there is no granular per-user consumption limit.

Diagnostic Steps 1 Check sys_gen_ai_usage_log : filter by date range → group by feature, user, skill → look for System Administrator or service account entries with massive counts. 2 Check sn_aia_execution_plan : filter by same date range → look for high-volume execution clusters on same records. 3 Check syslog for "Blocking AIA conversation" from sn_aia.AiAgentRuntimeUtilSNC — this indicates rate limiting is active. 4 Quick inventory of active triggers across all agents: query sn_aia_trigger_configuration filtered by active = true , and sn_aia_trigger_config_override filtered by active = true . Together these give the full list of trigger configurations currently armed across the platform — the fastest way to spot a runaway trigger before walking each workflow in Studio. 5 Review all active triggers: AI Agent Studio → each workflow → Define trigger → look for overly broad conditions. 6 Note: Usage data has a 1-day lag — data from March 1 appears on March 2. ℹ️ Important — these triggers are system events, not chat discovery. The trigger records on sn_aia_trigger_configuration and sn_aia_trigger_config_override are purely system events (record-driven, scheduled, or programmatic) and are NOT used by the chat discovery pipeline. A trigger firing here does not imply an NAP or VA user session — it is independent of the discovery chain documented in Map 2 above . Do not look for these executions in sys_cs_skill_discovery_tracking .

Resolution 1 Identify and deactivate runaway triggers immediately. 2 Disable auto-activated OOB skills that aren't needed. 3 Restrict NAP access by managing now_assist_panel_user role assignment. 4 For PRB1927995 (wrongly consumed assists): contact ServiceNow Support for assist consumption reversal. 5 Use Usage and Adoption dashboard to filter by product, channel, workflow, department. 6 Best practice: Never migrate triggers in active state. Set narrow trigger conditions. Monitor sys_gen_ai_usage_log weekly. Authoritative Rule — Trigger-Driven Surface Routing 📌 Trigger-driven conversations surface in whichever chat the user has access to (NAP or VA). If the user has access to neither, the agent or workflow must be designed to run fully autonomously — no interactive surface will be presented. Apply this rule when reviewing any trigger configuration:

User has NAP access (e.g., now_assist_panel_user role) → trigger-driven conversation surfaces in NAP. User has VA access only → trigger-driven conversation surfaces in Virtual Agent. User has neither (typical for service accounts, integration users, or system run_as users) → the agent / workflow has no interactive surface available . It must be authored to complete the task without any user prompt, confirmation, or input collection step. If the agent design includes interactive steps, those steps will fail or hang silently. Common authoring mistake: a trigger fires under a service-account run_as user with no NAP or VA access, but the agent contains an interactive "ask user to confirm" or "collect input" step. There is nowhere for that prompt to render — design the trigger path to be fully autonomous, or change the run_as to a user with NAP/VA access.

📘 KB2363662 — Admin Console Configuration 📘 KB2751618 — Assist Calculation 🔴 PRB1927995 — Wrongly consumed assists 📊 Use Case 12: Phantom Usage — Empty User & Skill Config ID in Analytics Logs Plugin Problem The User field and Skill Config ID on sys_gen_ai_usage_log are empty or inconsistent, breaking reports and dashboards. "Sum of AI Agent Assists Consumed" shows 0. Error: "insufficient access for 'query_range' operation on sys_gen_ai_usage_log.feature.feature_name."

Resolution 1 This is a known defect: PRB1952005 ("Empty users in sys_gen_ai_usage_log table"). 2 Update the Generative AI Controller ( sn_generative_ai ) to the latest version. 3 For access errors on the table: check user roles and table ACLs. 4 Note: sys_gen_ai_usage_log doesn't have separate properties for AI Agents vs Agentic Workflows — some features don't populate user details in the usage log (known limitation). 🔴 PRB1952005 — Empty users in usage log 🌐 Use Case 13: Multi-Language Meltdown — Non-English Agent DataType Failures & Dialect Drift I18N Plugin Problem AI Agents fail intermittently in non-English languages (especially Japanese). The agent picks incorrect DataTypes to display rather than collecting input from the user prompt. Two adjacent failure modes have also been observed at the response layer: dialect drift (BR-PT response on a PT-PT instance, or vice-versa) and citation strings remaining in English while the synthesized response is correctly translated.

Sub-Pattern A — DataType Failure (Japanese cluster) 1 This is a known issue tracked under PRB1975153. 2 Upgrade sn_aia to version 6.0+ for improved multi-language DataType handling. 3 If the issue persists, contact ServiceNow Support for the relevant fix. 4 For language translation in responses: check sn_generative_ai.append_lang_in_prompt — set to false for English-only responses regardless of input language. Sub-Pattern B — Dialect Drift (BR-PT vs PT-PT and similar) Reported on instances where both Brazilian Portuguese and European Portuguese were active simultaneously in Now Assist Admin → Languages, with the user preference set to PT-PT. The synthesized response came back in BR-PT. Removing BR-PT from active languages eliminated the prompt's language-instruction line but did not fix the response shape — the LLM (Azure OpenAI provider on the affected instances) was still selecting the wrong dialect. No workaround is currently available.

Multi-active-language ambiguity: Multiple active languages in Now Assist Admin combined with insufficient prompt instruction → LLM picks the closest dialect in its training distribution rather than the user's preference. Dynamic-translation toggle interaction: The interaction of sn_generative_ai.disable_dynamic_translation with the multi-language admin matrix is currently opaque. Capture the user-pref language, instance language, Now Assist Admin active languages, and the disable_dynamic_translation value when contacting Support. Sub-Pattern C — Citation Translation Regression NAVA renders the synthesized response in the user's language but the citation block (KB titles / source fragments coming back from FDIH) remains in English. Tracked under PRB2019099. No workaround is currently available; cluster-node restart does not resolve this issue. Multilingual citation regressions are also tracked under PRB2002666.

Anti-Patterns (do not repeat) Cluster node restart for translation cases — does not resolve the issue. Toggling glide.search.evam.use_legacy_template_generation for translation regressions — no effect (separate code path; this is tracked under PRB1850631 for that property). 🔴 PRB1975153 — Multi-language DataType handling (Japanese cluster) 🔴 PRB2002666 — Multilingual citations regression 🔴 PRB2019099 — Translation citation issue 🔘 Use Case 14: Agentic Support Toggle Won't Stick in VA Assistant Configuration NAVA Config Problem After plugin update, the "Agentic Support" toggle in Virtual Agent Assistant settings cannot be turned on. Saving has no effect — the toggle reverts to off. Console shows "Config.js: Attempted to access value before config loaded (key: FULFILLER_ENTITLEMENT)." Works for the default assistant but not for custom assistants.

Root Cause The fix script "Turn all Now Assist skills on" (sys_id: b0f0352cf7eb2210c4c99ef0e3bfd66f ) didn't add configuration attributes for custom assistants. Missing records "agentic_orchestration_enabled" and "use_agents_in_planner" in the sys_now_assist_deployment_config_attributes table.

Resolution 1 Navigate to sys_now_assist_deployment_config_attributes table. 2 Create TWO new records for the affected assistant: Record 1: attribute = "agentic_orchestration_enabled" (corresponds to "Use AI agent skills and orchestration" toggle). Record 2: attribute = "use_agents_in_planner" (corresponds to "Prioritize AI agents during skills discovery" checkbox). 3 Link both records to the affected assistant's sys_now_assist_deployment record. 4 The Agentic Support toggle should now function correctly. ✓ Related — inverse pattern: If the symptom is the opposite (the toggle silently flipped on after a plugin upgrade and previously-working topic flows started routing through the agentic pipeline unexpectedly), see UC19 below — that's the auto-flip pattern, not the won't-stick pattern, and the fix is different.

🎯 Use Case 15: Seven Deadly Patterns — Universal NAP & Trigger Failure Quick Reference NAP Trigger Config Purpose These seven patterns are common across NAP and trigger issues regardless of domain (ITSM, HR, CSM, FSM, Retail, GRC). Use this section as a quick reference when troubleshooting NAP or trigger problems.

Pattern A: Fulfiller Cannot See Now Assist Panel Check: User must have now_assist_panel_user role. Table: sys_user_has_role . Fix: Grant the role.

Pattern B: Agentic Workflow Not Triggering Automatically Checks: (1) AI Agent Studio → data trigger is enabled. (2) Record fields comply with trigger condition. (3) run_as field is NOT empty on the trigger. (4) run_as user has now_assist_panel_user role. Rule: a trigger-driven conversation surfaces in whichever chat the run_as user has access to (NAP or VA); if the run_as user has neither, the agent must be designed to run fully autonomously — no interactive surface is available. See the authoritative rule in UC 11 .

Pattern C: User Cannot Invoke Workflow from NAP Checks: (1) NAP display option toggled ON. (2) Setup roles match user's roles. (3) Agent is active ( sn_aia_agent.active = true ). (4) Agent proficiency is SET (empty = undiscoverable).

Pattern D: Execution Terminates at First Step Root Cause: Azure OpenAI connection not configured. Check: Connection & Credentials table → Azure OpenAI OEM connection. Verify URL points to llm-prod-proxy .

Pattern E: "Unable to Find Any Agents to Complete the Mission" Root Cause: Agent proficiency not set. Fix: Set proficiency on the sn_aia_agent record.

Pattern F: User Not Seeing AI Agent Triggered on NAP Checks: (1) run_as field on trigger is NOT empty. (2) run_as user has now_assist_panel_user role. (3) NAP is enabled: sys_now_assist_deployment.active = true .

Pattern G: Conversation Device Type Mismatch Table: sys_cs_conversation , field: device_type . Expected for AI Agent Studio testing: "Web Client". If NOT "Web Client" → check use case trigger: channel should be "Virtual Agent", portal should be set to desired portal. If consumer_account is wrong user → indicative of VA issue.

🧷 Use Case 16: Studio Skill-Config Binding Mismatch — config-id=undefined & "ACLs not defined" Banner Config Discovery Problem Inside AI Agent Studio, opening a skill that an agent uses shows a banner "ACLs are not defined" — yet opening the same skill independently via the standalone skill-kit URL with an explicit config-id shows that ACLs do exist on that skill. At runtime, the agent denies access ( "You don't have access to complete this request" ) at the first sub-agent or tool that has the binding gap, even though the user has the right roles. Customers find this perplexing because they SEE the ACLs when they open the skill independently.

Symptoms "You don't have the required access to complete this request. Try another request." — emitted mid-execution at a specific sub-agent or tool, not at the first step. Studio banner inside the skill detail view: "ACLs are not defined." The Studio URL for the skill contains config-id/undefined ; the standalone skill-kit URL has a populated config-id . ACLs are present in sys_security_acl filtered to the customer's app scope and active=true . Behavior is reproducible and not load-dependent. Root Cause The agent's sn_aia_agent_tool_m2m record points at one skill-config record while the active record on sys_gen_ai_skill for that prompt is a different one. The active record carries the ACLs; the stale record doesn't. The Studio surface deep-links via the m2m record (which is missing the config-id), producing the config-id=undefined URL and the misleading "no ACLs" banner. The agent runtime invokes the stale skill-config and is denied.

Diagnostic Steps 1 Reproduce in Studio: open the failing agent → Testing → run the workflow that fails. Note the sub-agent at which it terminates. 2 Compare URLs side-by-side. Open the failing skill from Studio (the URL will contain config-id/undefined ). Then open the same skill via the standalone skill-kit URL with the active config-id from sys_gen_ai_skill . If the standalone URL shows ACLs and Studio doesn't → binding mismatch confirmed. 3 Pull the agent's binding records: sn_aia_agent_tool_m2m filtered by the agent. For each tool, capture the bound skill_config sys_id. 4 Compare to the active skill-config: sys_gen_ai_skill filtered by the prompt; pull the active record's sys_id. 5 List ACLs: sys_security_acl filtered to the customer scope and active=true . Confirm the ACLs exist but are bound to the active config, not the one the m2m points at. Resolution 1 Rebind the agent's sn_aia_agent_tool_m2m record to the active skill-config sys_id from sys_gen_ai_skill . 2 Re-test from Studio; the "no ACLs" banner should clear and the runtime denial should resolve. 3 Any clone or customization of an OOB agent should re-run this binding audit — m2m records are scope-sensitive and clones frequently inherit stale binding. 💡 Field rule: Whenever a Studio banner contradicts a standalone skill-kit URL, suspect skill-config binding before suspecting ACL configuration. The diagnostic ladder of "user roles → agent Define-Data-Access → sub-agent Define-Data-Access → tool Role Masking → ACLs on the bound skill config" must be walked end-to-end on every clone.

📘 KB2521584 — AI Agents and Agentic Workflows Reference 📘 KB2507588 — Support reference for OOB AI Agents and skills 💳 Use Case 17: Custom Agents Hidden in Studio — Fulfiller Entitlement & AI-Native Tier Plugin Gating Config Plugin Problem Custom agents and agentic workflows do not appear in the AI Agent Studio create-and-manage page. The "New" button for agentic workflows is missing. The browser console shows "Unable to locate data broker with id = now_assist_for_platform_active, operation name = REFRESH" . The customer message reads "access limited due to license restrictions." Distinct from PRB-driven UI bugs because the diagnostic API explicitly returns false for the fulfiller artifact.

Diagnostic Script (run first) 1 Run in Background Scripts as admin: gs.info(new sn_entitlement.GlideEntitlementAPI().hasLicenseForArtifact("family_id","sn_gen_ai_platform_fulfiller",false,true)); 2 If output is false → continue with this UC. 3 If output is true but Studio still hides agents → fault is in the AI-Native tier plugin layer, not entitlement; jump to step 3 below. Three Disjoint Failure Modes Mode Cause Fix (i) True licensing gap Customer not entitled to the fulfiller SKU. AIA framework correctly denies. hasLicenseForArtifact returns false. Purchase the appropriate AI-Native tier to enable the entitlement. (ii) Plugin-install gap Fulfiller IS entitled but the tier plugins are missing — sn_ai_plat_fdn (Foundation), sn_ai_plat_adv (Advanced), sn_ai_plat_prm (Prime), sn_itom_ai_prime . UI hides custom agents/workflows. Install the matching tier plugin. Custom agents/workflows become visible only when Prime is installed; show/hide-advanced toggles depend on Foundation + Advanced + Prime. (iii) Cross-environment parity gap DEV has Now Assist licensed; PROD does not. sn_convo_studio , sn_vad_genai , Now Assist Skill Discovery and Execution missing in PROD. VAD Asset Library access is gated by Now-Assist roles + plugins PROD doesn't license. Plugin parity cannot be achieved if entitlements differ. Verify entitlement parity between environments before proceeding. Display-Name Sanity Check (rules out unrelated UI bug PRB2001660) 1 var p = new GlideRecord('sn_aia_property'); gs.info(p.getDisplayName()); 2 var r = new GlideRecord('sys_user_role'); gs.info(r.getDisplayName()); 3 var m = new GlideRecord('sn_aia_agent_tool_m2m'); gs.info(m.getDisplayName()); 4 All three should print "name" . Anything else → PRB2001660 (fixed in Now Assist AI Agents v7.1) is also in play. What Did Not Work (anti-patterns) Assuming this is purely a licensing question. The actual cause is sometimes a missing Prime plugin install rather than a licence gap. Always run the entitlement script AND the plugin-install check before concluding it is a licensing issue. Do not disable read ACLs on sn_aia_usecase or sn_aia_agent as a workaround for license gating. This is not a supported approach and can cause unintended access issues. ⚠️ Diagnostic check (recommended): Before investigating "custom agents missing in Studio" issues, run these three diagnostics: (1) hasLicenseForArtifact entitlement script, (2) Foundation/Advanced/Prime plugin presence, (3) display-name sanity check. The full picture allows for accurate root-cause identification.

📘 KB2507588 — ServiceNow Support Routing Reference 📘 KB2949962 — Known error for PRB2001660 🎮 Use Case 18: AI Agent Studio Test Button Greyed Out & Define User Access Dropdown 1000-Row Cap Config Discovery Problem Two related Studio surface issues that frequently arrive in the same case:

Test button stays greyed out in the Studio Testing page even after refresh; behaviour is intermittent — sometimes it enables, sometimes not. No console errors. "Define User Access" Role dropdown on sys_agent_access_role_configuration only shows the first 1,000 alphabetically-sorted roles (cuts off mid-alphabet); the role the customer needs is past the cap and not selectable from the picker. A related variant: Studio session expires prematurely while testing, especially in incognito browsers — points at an interaction between Studio's auth-cookie policy and the live-agent route used during testing. Root Causes (two threads) Studio UI binding bugs. The test-button enable check has a slow or race-conditioned dependency on data brokers. Refresh-and-retry sometimes works because it wins the race. Capture browser console, network HAR, and node logs at the moment of reproduction. Reference defects: PRB1962812. Skill-config-id=undefined when Studio links into skill-kit (see UC16 ). Fixing that binding fixes a whole family of "broken" Studio behaviours including the ACL banner and the test-flow short-circuit. Define User Access role-dropdown 1000-row cap. Sibling defect of PRB1978409 (which fixed the equivalent USERS dropdown but not the ROLES dropdown). The data broker that backs the roles picker has the same cap; for instances with more than 1,000 roles the customer's intended role isn't in the picker. Resolution 1 For the role-dropdown cap (workaround): Navigate to /sys_agent_access_role_configuration_list.do and edit the record directly via the list page. The persisted record has no 1,000-row constraint — only the picker does. 2 For an intermittent greyed test button: Verify Studio plugin ( sn_aia ) is on the latest available patch for the Glide family; clearing the cache after the upgrade has resolved several reported variants. 3 If Studio works inconsistently for a specific skill: Always open the same skill via standalone skill-kit URL with explicit config-id and compare. If standalone works and Studio doesn't, the agent is bound to a stale skill-config — see UC16. 4 For premature session expiry while testing: Capture HAR and node logs at the moment of expiry and contact ServiceNow Support. Re-testing in a non-incognito browser is a temporary workaround. 💡 Diagnostic approach: When a Studio issue appears intermittent, capture three things at the moment of reproduction before changing anything: browser console, network HAR, and the affected node's logs. These captures typically reveal a specific pattern even when behavior seems random.

🔴 PRB1978409 — Studio Define User Access USERS dropdown 1000-row cap (fixed); roles dropdown is a sibling defect 🔴 PRB1962812 — Studio Test button issues 🔁 Use Case 19: Agentic Orchestration Auto-Flips ON After Plugin Upgrade — Topic Flows Re-Routed Plugin Config NAVA Problem After an sn_nowassist_va plugin upgrade, agentic_orchestration_enabled flips silently from false to true on one or more Now Assist deployments. Assistants that previously routed through the classic topic flow now route through the agentic pipeline, breaking previously-working VA conversations. The customer never approved the flip. Distinct from UC14 (toggle won't stick / reverts to off) — this is the inverse: the toggle silently went on without action.

Symptoms One or more assistants suddenly start routing on the agentic pipeline post-upgrade. "Agentic Support" appears enabled "automatically" without customer action. Duplicate sys_now_assist_deployment_config_attributes records for agentic_orchestration_enabled on the same assistant — one in the customer scope, one inserted by an upgrade-bundled update set. Symptom presents only on instances upgraded via plugin install/update; not on instances that received the same code via update-set movement. Root Cause The upgrade ships a fix script that flips agentic_orchestration_enabled to true on every deployment whose config attribute ai_agent_skill_enabled is true at upgrade time. That predicate is the customer-actionable hook for prevention (see Option A below). The fix script only runs on plugin install/update; it does not run on update-set movement. That asymmetry is what produces the "works fine in lower environments where we moved updates, broken in environments where we ran the upgrade" pattern.

Resolution Options 1 Option A — pre-upgrade prevention: Before the plugin upgrade, set ai_agent_skill_enabled = false on the sys_now_assist_deployment_config_attributes record for each assistant that should NOT be flipped. The fix script's predicate fails and the flip is skipped. Revert to true after upgrade if desired. 2 Option B — post-upgrade rollback (most commonly used): Let the upgrade run; immediately afterwards, set agentic_orchestration_enabled = false on the affected deployments. Confirmed to restore classic topic-flow routing. 3 Audit step in either case: Query sys_now_assist_deployment_config_attributes grouped by deployment + attribute name. Look for duplicates of agentic_orchestration_enabled — when an upgrade-bundled update set inserts a second record, only one of the two will be honoured at runtime depending on scope-resolution order. De-duplicate. Pre-Upgrade Validation Checklist (to run on every Now Assist plugin upgrade) 1 List every assistant's current agentic_orchestration_enabled and ai_agent_skill_enabled values from sys_now_assist_deployment_config_attributes . 2 Decide deployment-by-deployment whether the upgrade-time flip is desired. 3 For deployments where it is NOT desired, apply Option A above before the upgrade. ⚠️ Why this is easy to miss: Customers who promote Now Assist via update sets rather than plugin upgrades will not see the flip in their lower environments — the predicated fix script never runs. The first time the flip occurs is in the environment where the actual plugin install/update happens. Document the upgrade-time behaviour in the change record so it isn't a surprise in PROD.

📘 KB2913453 — How to revert Agentic Support for an assistant ⚙️ Key Plugins & Properties Reference Required Plugins & Recommended Versions Plugin / App App ID Purpose & Notes Now Assist Core com.now_assist_core Core framework — repair first in any plugin repair sequence Generative AI Controller sn_generative_ai LLM proxy, Gen AI logging, Role Masking — upgrade for PRB1952005 fix. Never install standalone. Platform AI Agents and Skills sn_uxc_gen_ai Platform skills and agent framework. v11.0 fixes PRB1908380, PRB1962178. v11.5.4 on ZP6. Now Assist AI Agents sn_aia Agent orchestration. v5.2+ for field hierarchy; v6.0+ for multi-language; v6.1.5+ for isNextExecutionModeRoute fix; v7.0+ for parent-table fields and PRB1986221 NASDE (Skill Discovery) sn_vad_genai Skill discovery & execution in VA. Outdated versions cause "Synthesized Result" bugs. Key versions: 7.1.9, 7.2.22, 8.0.13 Now Assist Admin Console sn_nowassist_admin Admin UI — repair last in sequence. v6.6.4+ enforces Fulfiller SKU checks AI Search RAG sn_ais_rag RAG Retrieval tools. v5.0.2+ required for recent release compatibility Now Assist in Virtual Agent sn_nowassist_va VA integration layer NAVA Configuration sn_nava_config NAVA settings. Key versions: 6.0.2, 7.0.14 Omni-Experience sn_oe_sfs UX framework — outdated version blocks Conversational Studio AI Engagement Experience sn_uxc_gen_ai (1.0.6) Fixes blank NAP on Zurich P1. Houses com.glide.agentic_processes_view.enabled property Identity Agent Security com.glide.identity.agent.security Controls role masking behavior — its presence changes how strictly masking is enforced Key System Properties Property Purpose glide.cache.size.SkillDetailsCache Controls skill details cache size. Increase for instances with many skills (e.g., 450+) to avoid slow discovery. sn_vad_genai.com.glide.cs.one_extend.auto_proxy_enabled Enables auto-proxy for One Extend calls. Should be true . If missing → VA fails. sn_generative_ai.append_lang_in_prompt Appends user language to LLM prompt. Set to false for English-only responses. sn_vad_genai.input_collector.executor Controls input collector execution mode. Toggle between "New Data Collector" and "Input Collection Execution" to resolve input failures. glide.cs.runtime.user_input_max_length Maximum user-input length passed to runtime. OOB 500000 . May need to be increased to 5000000 for very large summarization inputs. com.glide.cs.fdih.interactive.timeout Global per-tool FDIH interactive timeout. Per-tool override is on sn_aia_agent_tool_m2m.timeout . Direct edits to one_api_service_plan_feature.timeout_sec revert (KB1528150). Key Diagnostic Tables Table Purpose sys_generative_ai_log Primary log for all Gen AI / LLM interactions. Filter by conversation sys_id. Prompt field has restricted access. sys_cs_skill_discovery_tracking Records which skills were discovered. Empty Available Skill IDs = discovery failure. sys_cs_fdih_invocation FDIH handler. Missing records = request never reached the AI framework. sys_cs_context_profile Assistant context profile. Inactive = blocks all NAP discovery. Synced via BR on sys_now_assist_deployment. sys_gen_ai_skill Skill registration. Missing records prevent NAP activation. Orphaned records (empty skill_document) cause cache overflow. sn_aia_skill_metadata Agent skill metadata. Must be in sync with sys_gen_ai_skill . sys_gen_ai_usage_log Usage analytics. Empty User field indicates PRB1952005. 1-day lag on data. sys_upgrade_history Verify store app upgrades were actually applied after platform upgrade. sys_gen_ai_feature_mapping LLM feature mapping. AIA ReAct Engine record must be present for NowLLM. sys_now_assist_deployment_config_attributes NAP config attributes. Must contain agentic_orchestration_enabled and use_agents_in_planner for custom assistants. sys_cs_channel Channel config. supports_multiple_conversations flag can cause NAP to not load (PRB1965617).