Skip to main content
Getting started

Setup wizard: AI and Q-Director

Steps 4 to 9 of the setup wizard: choosing managed AI or your own keys and models, then setting Q-Director's analysis, triage, intelligence, web search and Customer Success options, with safe choices for a first setup.

Written By Chris Scaminaci

Last updated About 3 hours ago

Steps 4 to 9 of the setup wizard decide how AI runs and how Q-Director, the part of QuantumOps that analyses and triages tickets, behaves on your desk. Only an Administrator can use them. Every setting here can be changed after setup, so a first setup does not have to be perfect.

Before you start: finish steps 2 and 3. The wizard unlocks these steps in order.

Safe choices for a first setup

StepA sensible first choice
4Managed AI: leave the key fields empty, keep the models the wizard preselects, keep Smart Queuing on.
5Keep the starting values. Where the screen shows a recommended value, use it. Leave category creation off.
6Leave Post Sherlock Results to Ticket, auto-triage, on-call alerts and escalation paths off until you have read the page that covers each.
7Keep the starting values. Leave the dispatcher workflow off.
8Keep Tavily + Bing.
9Switch the dashboard on only if you plan to use Customer Success.

Some options change things in HaloPSA: Allow Q-Director to Create New Categories, Re-categorize on Ticket Closure and Set Resolution Category on Closure in step 5, Post Sherlock Results to Ticket in step 6 and Halo Colour Sync in step 7, which sets the colour of tickets. Switch each of them on deliberately, not by accident. The triage options in step 6 write nothing to HaloPSA by themselves: dispatchers still triage and assign tickets in HaloPSA. See Q-Director: triage and routing.

Step 4: AI Configuration

The header contrasts two ways to run AI: Managed (pay per use) and BYOK (your API keys). Leave the API Key under Anthropic empty to use managed AI. Enter your own keys to bring your own.

Choose the models

Select a model field to open the model picker. Models from providers you have not switched on are greyed out.

FieldWhat it is used for
Primary ModelRequired. Bulk ticket analysis, categorisation and sentiment.
Secondary ModelTriage, incremental refresh and quick checks. A fast model suits it.
Reasoning ModelComplex planning and deep analysis.
Large Context ModelOperations over 200K tokens, such as memory analysis and profile generation. Leave it empty and QuantumOps chooses from your primary model. Only models with extended context are eligible.
Sanity Check Provider and Sanity Check ModelThe provider (OpenAI (GPT), Anthropic (Claude) or Together.ai) and model that cross-check results. Leave the model empty for the provider's default.

Providers

  • Anthropic (Claude) has a switch. The API Key is optional. Under Rate Limiting, set the Service Tier to match your Anthropic account: the choices run from Tier 1 - Free/Starter to Monthly Invoicing (Highest), and the line under the field shows the token and request limits QuantumOps will pace to. A tier higher than your account has makes requests fail when Anthropic's limit is reached. With managed AI, keep the preselected tier. Timeout (min) takes 1 to 15 minutes. Smart Queuing queues requests when a limit is hit and is recommended.
  • OpenAI is optional. Switch it on and enter an OpenAI API Key if you want GPT models as a fallback.
  • Together.ai is optional. Switch it on to use open-source models such as Llama, DeepSeek and Qwen. It needs your own Together.ai API Key. Token Cap/Min, Request Cap/Min and Timeout (min) set its limits.
  • Vector Store offers Managed Vector Store, which needs no configuration, and Managed Embeddings runs automatically. There is nothing to set.

Next needs a Primary Model and at least one provider switched on, and a key if you switched on Together.ai.

Later, change these settings in the AI Provider Configuration section of Tenant settings. AI providers and models explains the choices in depth.

Step 5: Q-Director Core

Step 5 sets the foundations. Every field starts with a working value, so you can continue without changing anything. Under most numeric fields the wizard prints a recommended value and the reason for it.

  • Incremental Lifecycle Analysis. Enable Incremental Refreshes analyses a ticket again when it changes, so insights stay current without running a full reprocessing job. When it is on you also see AI Model, which is your Secondary Model, Stale Recheck (hours), Scan Interval (minutes) and Timeline Retention (days).
  • Ticket Categorization. Enable AI Ticket Categorization lets AI assign categories. Categorization Timing chooses when: OnCreation, DuringTriageAssist or OnClosure. Allow Q-Director to Create New Categories lets it create a category in HaloPSA when none of your existing ones fits, and Category Creation Mode chooses between Category, ResolutionCategory and Both. Re-categorize on Ticket Closure and Set Resolution Category on Closure update the category when a ticket is resolved.
  • Stale Ticket Detection. These settings decide when a ticket counts as stale, and they drive health scores and alerts.
    • Inactivity Days is the number of business days without activity, and Max Ticket Age (days) is the age after which a ticket counts as too old. Count Business Days Only skips weekends. Inactivity Trigger chooses whose silence matters: NoAgentAction, NoUserAction, NoAnyAction or AgentAwaitingResponse.
    • Track Agent Follow-Up and Follow-Up Days flag tickets where an agent has not followed up in time.
    • Assignment Delay (hrs) is how long an unassigned ticket can wait. SLA Hold Time (hrs) is the threshold for tickets on SLA hold. Track SLA Breaches Separately treats a breach as more serious than staleness.
    • Enable AI Stale Analysis asks AI why a ticket is stale. Analysis Trigger chooses when: IncrementalRefresh, ScheduledOnly or Disabled. Min Interval (hrs) sets the minimum gap between analyses of the same ticket.

Later, find these in Q-Director Settings: incremental refresh and stale detection on the AI Processing tab, categorisation on the Classification tab. Some labels differ there; Find the same settings in Q-Director Settings maps them. See Q-Director Settings, Q-Director: AI Processing, Q-Director: sentiment and categorisation and Stale ticket detection, which also gives the default for each stale setting.

Step 6: Q-Director Automation

  • Triage Assist. Enable Triage Assist gives dispatchers AI suggestions for routing and prioritisation. Set Impact & Urgency During Triage sits under it. QuantumOps does not write impact and urgency to a ticket on its own: dispatchers set them with Dispatch on the Ticket Triage Dashboard.
  • Research & Sherlock. Post Sherlock Results to Ticket adds Sherlock's research findings to the ticket in HaloPSA as an action. It is a choice you make, so decide it deliberately. Use Custom Action Filtering lets you list HaloPSA action outcomes to leave out of AI analysis in Excluded Outcomes (comma-separated), for example Rule Applied or SLA Hold. See Sherlock research and live context.
  • Triage Dashboard. Enable Triage Dashboard shows the triage dashboard's settings (duplicates, recurring incidents, feedback and display, which step 7 holds) in Q-Director Settings. The Ticket Triage Dashboard that dispatchers open from a HaloPSA tab opens whether it is on or off. See The Ticket Triage Dashboard and Q-Director: triage and routing.
  • Auto-Triage & Assignment. Enable Auto-Triage & Assign has Idle Time Before Auto-Triage (minutes), Schedule Time Zone and Triage Outcome ID. The outcome ID is the number of the HaloPSA action outcome your team logs once a ticket has been triaged. A ticket update that carries it marks triage as completed so the ticket moves on in Dispatch. Turning the switch on does not make QuantumOps triage or assign tickets by itself; see what Auto triage and assign tickets does. Q-Director Settings lets you pick the outcome by name instead of typing its number.
  • On-Call Notifications. Notify On-Call Staff has a Notification Type of PagerDuty, SMS or Email, and a target to match.
  • Escalation Paths. Enable Escalation Paths is a switch here. The detailed configuration is in Q-Director Settings after setup.

Leave auto-triage, on-call notifications and escalation paths off in a first setup, and read Q-Director: triage and routing before you switch any of them on.

Later, find these on the Triage & Routing tab of Q-Director Settings. Sherlock posting is on the Research & Docs tab, and custom action filtering is on the AI Processing tab. Some labels differ there; see Find the same settings in Q-Director Settings.

Step 7: Q-Director Intelligence

  • Sentiment & Tonality Analysis. Analysis Sensitivity is Low, Medium or High: low gives fewer alerts, high is more vigilant. Sentiment Alert Threshold (1-10) and Tonality Alert Threshold (1-10) raise an alert when the score is at or below the number. Enable Sentiment Feedback Loop adds feedback buttons to Slack and Teams alerts. Custom Analysis Instructions is optional text that tailors the analysis to your business. See Sentiment and tonality alerts.
  • Duplicate Detection. Detection Scope is Client, Global or User. Lookback Days sets how far back to search, Ticket Status Filter chooses OnlyOpen, OpenAndClosed or OpenAndRecentlyClosed, and Recently Closed Days appears for the last of those. Similarity Threshold is a slider. Subject Line Matching, Auto-Flag Duplicates and Show Merge Button are checkboxes. Duplicate tickets and recurring issues explains the thresholds.
  • Recurring Incident Detection. Enable Recurring Detection has a threshold for User, Client and Global, and a Window (days).
  • ML Feedback & Learning. Enable ML Feedback Loop is the switch for the feedback loop that works from dispatcher corrections. Under it, Require Manual Review of Corrections asks for human approval before learned changes apply, and Min Corrections Before Retraining sets how many corrections to collect first. When you change Require Manual Review of Corrections, the wizard also sets manager review in the dispatcher workflow. See Dispatch workflow and scheduling settings. See Override and feedback review.
  • Dashboard Display Settings. Evaluation Mode is Auto, Manual or Hybrid. Refresh Interval (sec) and Max Similar Tickets to Show are numbers. Agent Workload, Client Health Score, Priority Score and Preload Predictions choose what the triage dashboard shows.
  • Dispatcher Workflow. Enable Dispatcher Workflow turns on dispatcher-focused scheduling, routing and guided work. Appointment Scheduling and Halo Colour Sync are checkboxes, and colour sync sets ticket colours in HaloPSA. Workflow Rollout Phase runs from Disabled through RecommendOnly, DispatcherApprovalRequired, ManagerApprovalRequiredForLearning and ColourSyncEnabled to AutoSchedulingEligible. Keep the switch off and the phase on Disabled until you are ready to roll the workflow out. See Dispatch workflow and scheduling settings.

Later, find sentiment on the Classification tab of Q-Director Settings, duplicates, recurring incidents, feedback and dashboard display on the Triage & Routing tab, and the dispatcher workflow on the Dispatch & Scheduling tab. The setup page for each is Q-Director: sentiment and categorisation and Q-Director: triage and routing.

Find the same settings in Q-Director Settings

The wizard and Q-Director Settings label some settings differently. Where a label in steps 5 to 7 differs, this table gives the label to look for in Q-Director Settings. A setting that is not listed has the same label in both places.

StepIn the wizardIn Q-Director Settings
5Enable Incremental RefreshesEnable incremental refreshes on ticket updates
5Scan Interval (minutes)Scan Interval (mins)
5Inactivity DaysInactivity Threshold (business days)
5Track Agent Follow-UpTrack Agent Follow-Up Requirements
5Follow-Up DaysFollow-up Required Within (days)
5Assignment Delay (hrs)Assignment Delay (hours)
5SLA Hold Time (hrs)SLA Hold Time Threshold (hours)
5Min Interval (hrs)Min Hours Between Analysis
5Enable AI Ticket CategorizationEnable Q-Director ticket categorization
5Categorization TimingWhen to categorize
5Allow Q-Director to Create New CategoriesAllow Q-Director to create categories
5Re-categorize on Ticket ClosureRecategorize ticket on closure
5Set Resolution Category on ClosureSet resolution category on closure
6Enable Triage AssistEnable Q-Director Triage Assist
6Enable Triage DashboardEnable Triage Dashboard for dispatchers
6Enable Auto-Triage & AssignAuto triage and assign tickets
6Idle Time Before Auto-Triage (minutes)Auto triage after idle time (minutes)
6Triage Outcome IDTriage outcome to use
6Set Impact & Urgency During TriageSet Impact and Urgency during auto triage
6Post Sherlock Results to TicketPost Sherlock research results to ticket
6Use Custom Action FilteringUse custom action filtering for AI analysis
6Notify On-Call StaffNotify on-call staff of after hours tickets
7Analysis SensitivitySensitivity Level
7Enable Sentiment Feedback LoopEnable Feedback Buttons in Alerts
7Lookback DaysLookback Period (days)
7Ticket Status FilterStatus Filter
7Recently Closed DaysRecently Closed Period
7Similarity ThresholdContent Similarity Threshold
7Subject Line MatchingEnable subject line matching
7Auto-Flag DuplicatesAuto-flag high-confidence duplicates
7Show Merge ButtonShow merge button
7Enable Recurring DetectionEnable recurring incident detection
7Enable ML Feedback LoopTrack dispatcher corrections
7Require Manual Review of CorrectionsEnable Manager Review, on the Dispatch & Scheduling tab
7Refresh Interval (sec)Auto-refresh (seconds)
7Max Similar Tickets to ShowMax Similar Tickets
7Preload PredictionsPre-generate AI predictions

Two settings keep their label but offer different choices. Detection Scope offers Client, Global or User in the wizard and Same Client, Same User or Global in Q-Director Settings. Evaluation Mode offers Auto, Manual or Hybrid in the wizard and Quick, Deep or Auto in Q-Director Settings. Stale ticket detection gives the default for each stale setting.

Web search lets AI look things up while it analyses a ticket: error codes, documentation, known issues and unfamiliar technologies. Choose a Search Provider:

  • Tavily + Bing is the recommended managed service. It needs no setup.
  • Anthropic Web Search is Claude's own web search. It needs your own Anthropic API key with web search enabled on the account. Choosing it switches on Enable Anthropic Web Search and shows Max Searches Per Request (1 to 20, with 3 to 5 recommended), and the optional Allowed Domains and Blocked Domains, each a comma-separated list. Leave Allowed Domains empty to allow all domains.

Later, change these in Q-Director: web search.

Step 9: Customer Success

Customer Success monitors the health of your clients and tracks satisfaction surveys.

  • Enable Customer Success Dashboard switches the feature on. The options below appear once it is on.
  • Rating Scale is fixed: surveys are rated 1 to 5 with faces. You choose 3 or 5 faces later in CSAT Administration → Survey Config.
  • Link Expiration (days) is how long a survey link keeps working.
  • Post CSAT Results as Private Actions adds feedback to the ticket as private notes that customers cannot see.
  • Alert Silence Duration (hours) is how long to silence an alert after someone acknowledges it.

Later, see Customer Success settings and CSAT Administration.

What comes next

Step 10 shows the addresses for your HaloPSA tabs. See Adding QuantumOps tabs to HaloPSA. Step 11 reviews your choices and completes setup. See The setup wizard.